https://www.na-mic.org/w/api.php?action=feedcontributions&user=Serdar&feedformat=atomNAMIC Wiki - User contributions [en]2024-03-28T12:39:12ZUser contributionsMediaWiki 1.33.0https://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=275912008 Summer Project Week:GroupwisePortingToNamicKit2008-06-26T15:38:57Z<p>Serdar: /* Key Investigators */</p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|<br />
|[[Image:GroupwiseSummary.PNG|320px]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Utah: Casey Goodlett<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
* Extended file extension support, fixed bugs about transform file outputs.<br />
* Added verbose messages to registration tool.<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
===References===<br />
[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| Groupwise Regstration - NA-MIC Publications Database]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=275862008 Summer Project Week:GroupwisePortingToNamicKit2008-06-26T15:26:16Z<p>Serdar: /* Key Investigators */</p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|<br />
|[[Image:GroupwiseSummary.PNG|320px]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Utah: Casey Goodlett<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
Fixed several bugs about the use of the code.<br />
Added more verbose outputs to registration tool.<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
===References===<br />
[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| Groupwise Regstration - NA-MIC Publications Database]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=273502008 Summer Project Week:GroupwisePortingToNamicKit2008-06-23T17:31:59Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|<br />
|[[Image:GroupwiseSummary.PNG|320px]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Utah: Casey Goodlett<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
Currently, we are pre-processing the clinical dataset.<br />
<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| Groupwise Regstration - NA-MIC Publications Database]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=273102008 Summer Project Week:GroupwisePortingToNamicKit2008-06-23T15:57:06Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|<br />
|[[Image:GroupwiseSummary.PNG|320px]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
Currently, we are pre-processing the clinical dataset.<br />
<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| Groupwise Regstration - NA-MIC Publications Database]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=273062008 Summer Project Week:GroupwisePortingToNamicKit2008-06-23T15:54:31Z<p>Serdar: /* References */</p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
Currently, we are pre-processing the clinical dataset.<br />
<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| Groupwise Regstration - NA-MIC Publications Database]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week:GroupwisePortingToNamicKit&diff=273012008 Summer Project Week:GroupwisePortingToNamicKit2008-06-23T15:48:32Z<p>Serdar: New page: {| |Project Week Main Page ]] |} __NOTOC__ ===Key Investigators=== * MIT: Serdar Balci, Polina Golland *...</p>
<hr />
<div>{|<br />
|[[Image:ProjectWeek-2008.png|thumb|320px|Return to [[2008_Summer_Project_Week|Project Week Main Page]] ]]<br />
|}<br />
<br />
<br />
__NOTOC__<br />
<br />
<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Brad Davis<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
We aim to make groupwise registration code publicly accessible, by integrating the code into the NAMIC-kit.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan</h1><br />
Our approach is to provide a complete package including the code and a test dataset.<br />
We already presented the algorithm in MICCAI-2007 open-source workshop and submitted the code to NAMIC-Sandbox.<br />
<br />
During the project week, we plan to provide a clinical dataset for experimental evaluation of the groupwise registration algorithm.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
Currently, we are pre-processing the clinical dataset.<br />
<br />
<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
* Serdar K. Balci, Polina Golland, William M. Wells. Non-rigid Groupwise Registration using B-Spline Deformation Model. MICCAI Workshop on Open Source and Open Data 2008.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week&diff=272892008 Summer Project Week2008-06-23T15:26:36Z<p>Serdar: /* Other Projects */</p>
<hr />
<div>Back to [[Engineering:Programming_Events|Programming/Project Events]]<br />
<br />
[[Image:ProjectWeek-2008.png|thumb|220px|right|Summer 2008]]<br />
<br />
== Logistics ==<br />
*'''Dates:''' June 23-27, 2008<br />
*'''Location:''' MIT. [[Meeting_Locations:MIT_Grier_A_%26B|Grier Rooms A & B: 34-401A & 34-401B]].<br />
*'''Registration Fee:''' $260 (covers the cost of breakfast, lunch and coffee breaks for the week). Due by Friday, June 13th, 2008. Please make checks out to "Massachusetts Institute of Technology" and mail to: Donna Kaufman, MIT, 77 Massachusetts Ave., 38-409a, Cambridge, MA 02139. Receipts will be provided by email as checks are received. Please send questions to dkauf at mit.edu. If you are attending for one day only, the registration fee is not required.<br />
*'''Registration Method''' The event is full. [[User:Tkapur|Tkapur]] 18:12, 17 June 2008 (UTC)<br />
*'''Hotel:''' We have a group rate of $239/night (plus tax) for a room with either 1 king or 2 queen beds at the [http://www.hotelatmit.com Hotel at MIT (now called Le Meridien)]. [http://www.starwoodmeeting.com/StarGroupsWeb/booking/reservation?id=0805167317&key=4FD1B Please click here to reserve.] This rate is good only through June 1.<br />
*Here is some information about several other Boston area hotels that are convenient to NA-MIC events: [[Boston_Hotels|Boston_Hotels]]. Summer is tourist season in Boston, so please book your rooms early.<br />
*2008 Summer Project Week [[NA-MIC/Projects/Theme/Template|'''Template''']]<br />
*[[2007_Programming/Project_Week_MIT#Projects|Last Year's Projects as a reference]]<br />
*For hosting projects, we are planning to make use of the NITRC resources. See [[NA-MIC_and_NITRC | Information about NITRC Collaboration]]<br />
*Next Project Week in Utah -- January 5-9, 2009<br />
<br />
==Introduction to NA-MIC Project Week==<br />
Please read an introduction about these events [[Project_Events#Introduction|here]].<br />
<br />
== Agenda==<br />
* Monday <br />
** noon-1pm lunch <br />
**1pm: [[2008-Project-Week-Welcome|Welcome]] (Ron Kikinis)<br />
** 1:05-3:30pm Introduce [[#Projects|Projects]] using templated wiki pages (all Project Leads) ([[NA-MIC/Projects/Theme/Template|Wiki Template]]) <br />
** 3:30-5:30pm Start project work<br />
* Tuesday <br />
** 8:30am breakfast<br />
** 9:00-9:45am: NA-MIC Software Process <br />
** 10-10:30am [http://www.slicer.org/slicerWiki/index.php/Announcements:Slicer3.2 Slicer 3.2 Update] (Jim Miller, Steve Pieper)<br />
** 11-12noon [[Project Week 2008 Slicer Tuning| Performance tuning for Slicer 3.2]] (Jackson Room 38-466) (Ron Kikinis)<br />
** noon lunch<br />
** 1pm: [[2008-Project-Week-SVN-Change|Cut over to a new version of the NA-MIC SVN]]. (Grier Room) Zack Galbreath will provide details and can help with configurations.<br />
** 2:30-3:30pm: [[Project Week 2008 Special topic breakout: XNAT Database]] (Stata 32-D407) (Daniel Marcus) <br />
** 5:30pm adjourn for day<br />
* Wednesday <br />
** 8:30am breakfast<br />
** 9:00-12pm [[Project Week 2008 Special topic breakout: ITK]] (Luis Ibanez)<br />
** noon lunch<br />
** 2:30-3:30pm: [[Project Week 2008 Special topic breakout: Non-rigid Registration]] (Stephen Aylward)<br />
** 5:30pm adjourn for day<br />
* Thursday<br />
** 8:30am breakfast<br />
** noon lunch<br />
**2:30-3:30pm [[Project Week 2008 Special topic breakout: GWE]] (Marco Ruiz)<br />
** 5:30pm adjourn for day<br />
* Friday <br />
** 8:30am breakfast<br />
** 10am-noon: Project Progress using update [[#Projects|Project Wiki pages]]<br />
** Noon: Lunch boxes and adjourn. (Next one [[AHM_2009| in Utah the week of Jan 5, 2009]])<br />
<br />
== Projects ==<br />
<br />
===DBP II===<br />
These are projects by the new set of DBPS:<br />
#[[DBP2:Harvard|Velocardio Facial Syndrome (VCFS) as a Genetic Model for Schizophrenia]] (Harvard: Marek Kubicki, PI)<br />
##[[2008_Summer_Project_Week:DWIRegistrationOMT|DWI Registration using Optimal Mass Transport]] (Sylvain Bouix BWH, Tauseef Rehman GATech)<br />
##[[2008_Summer_Project_Week:EddyCurrentCorrection|EPI-DWI Eddy Current distortion correction]] (Sylvain Bouix BWH, Ran Tao Utah)<br />
##[[2008_Summer_Project_Week:LobeParcellation| Parcellation of 3T MR data]](Sylvain Bouix BWH, Priya Srinivasan BWH, Brad Davis Kitware)<br />
##[[2008_Summer_Project_Week:GMLongDistanceTractography|GM Long Distance Tractography]] (John Melonakos GATech, Marek Kubicki BWH)<br />
##[[2008_Summer_Project_Week:PopulationDTIApplication|Group Analysis of DTI]] (Casey Goodlett Utah, Marek Kubicki BWH)<br />
#[[DBP2:UNC|Longitudinal MRI Study of Early Brain Development in Autism]] (UNC: Heather Hazlett, Joseph Piven, PI)<br />
##[[2008_Summer_Project_Week:RegionalCorticalThicknessTool|Work Flow Tool for regional cortical thickness pipeline]] (Clement Vachet UNC)<br />
##[[2008_Summer_Project_Week:NITRCRegistration|NITRC registration of cortical thickness modules]] (Clement Vachet UNC)<br />
##[[2008_Summer_Project_Week:DWI-DTI_PrepTools|DWI DTI Prep Tools]] (Zhexing Liu UNC) <br />
#[[DBP2:MIND|Analysis of Brain Lesions in Lupus]] (MIND/UNM: Jeremy Bockholt, Charles Gasparovic PI)<br />
##[[DBP2:MIND:RoadmapProject|Lesion Classification Module]] (Mark Scully MIND)<br />
##[[DBP2:MIND:LongitudinalRegistrationProject|Longitudinal Registration and time course analyses in white matter lesions]] (Mark Scully and Jeremy Bockholt, MIND)<br />
##[[DBP2:MIND:BeyondLesionsProject|Enhancements and extension of white matter lesion classification using DTI scalars]] (Jeremy Bockholt, MIND)<br />
#[[DBP2:JHU|Segmentation and Registration Tools for Robotic Prostate Intervention]] (Queens/JHU: Gabor Fichtinger, PI)<br />
##[[2008_Summer_Project_Week:TransRectal_Prostate_Biopsy_Module|Trans-Rectal Prostate Biopsy module]] (David Gobbi, Gabor Fichtinger, Queens/JHU)<br />
##[[2008_Summer_Project_Week:ProstateSegReg|Prostate Segmentation and Registration]] (Yi Gao GATech, Gabor Fichtinger JHU)<br />
##[[2008_Summer_Project_Week:PerkStation|Hardware/software overlay for percutaneous intervention (PERK Station)]] (Siddharth Vikal, Gabor Fichtinger, Queens/JHU)<br />
<br />
===Other Projects===<br />
#[[2008_Summer_Project_Week:EddyCurrentCorrection|Eddy current and head motion correction of DWIs]] (Ran Tao, Utah, Sylvain Bouix, BWH, Xiaodong Tao, GE, Tom Fletcher, Utah)<br />
#[[2008_Summer_Project_Week:GroupwiseBSplineForDTI| Integraton of groupwise b-spline registration into atlas building]] (Casey Goodlett, Serdar Balci)<br />
#[[2008_Summer_Project_Week:GroupwisePortingToNamicKit| Porting groupwise registration project into NAMIC-kit]] (Serdar Balci, Brad Davis)<br />
# [[2008_Summer_Project_Week:CVS_SVN_Synchronization|CVS / SVN auto synchronization]] (Sebastien, Steve, Jim, Will, Bill)<br />
# [[2008_Summer_Project_Week:3DWidgetsInSlicer|3D Widgets in Slicer]] (Nicole Aucoin, Will Schroeder)<br />
## Issues with existing widgets<br />
## Design of new widgets<br />
# [[2008_Summer_Project_Week:Batch_Processing|Batch processing in the NAMIC Kit]] (Julien, Marco, Steve, Jim)<br />
#[[2008_Summer_Project_Week:ModuleChaining|Module Chaining]] (Marco, Jim, Steve, Dan B., Luca)<br />
# [[2008_Summer_Project_Week:Nonlinear transforms | Nonlinear transforms]] (Jim, Steve, Luis)<br />
## TransformToWorld/TransformFromWorld, integration with slice viewing<br />
# [[2008_Summer_Project_Week:XNATandXCEDE| Slicer3, XNAT integration and XCEDE Web Services ]] (Dan M., Wendy, Steve, Julien, Dan B.)<br />
## Review and enrich use cases [[Media: XCEDE-Use-Cases-2008-06-25.ppt | (developing use case ppt)]]<br />
# [[2008_Summer_Project_Week:PythonInSlicer| Python in Slicer]] (Dan B., Michael Halle, Steve, Luca)<br />
# [[2008_Summer_Project_Week:PerformanceTuningFiducials|Performance Tuning of Fiducials]] using the EventBroker and other tools (Nicole Aucoin, Alex Yarmarkovich, Steve Pieper, Will Schroeder)<br />
# [[2008_Summer_Project_Week:FocusedGUIRefinement|Focused GUI Refinement and Strategies for Consistency]] (Wendy, Sebastien) [http://www.na-mic.org/Bug/view.php?id=242]<br />
# [[2008_Summer_Project_Week:fMRIconnectivity|fMRI connectivity]] (Bryce Kim, MIT)<br />
# [[2008_Summer_Project_Week:AtlasFreeSegmentation|Atlas free Segmentation]] (Tammy Riklin-Raviv, MIT)<br />
#[[NA-MIC/Projects/Collaboration/EM Bias Field Correction | New Bias Field Correction in EM]] (Carlos Sánchez Mendoza, Kilian Pohl - SPL, Brad Davis - Kitware)<br />
# [[2008_Summer_Project_Week:FluidMechanicsTractographyUCLA|Fluid Mechanics Based DTI Tractography]] (Nathan Hageman, UCLA)<br />
<br />
===External Collaborations===<br />
#[[NA-MIC/Projects/Collaboration/UWA-Perth]] (Adam Wittek)<br />
#[[NA-MIC/Projects/Collaboration/MRSI Module for Slicer]] (Bjoern Menze)<br />
#[[NA-MIC/Projects/Collaboration/NIREP: Non-rigid Image Registration Evaluation]] (Gary Christensen Group)<br />
#[[NA-MIC/Projects/Collaboration/Lung Atlas]] (Gary Christensen Group)<br />
#[[NA-MIC/Projects/Collaboration/Non-rigid image registration]] (Gary Christensen Group)<br />
#[[NA-MIC/Projects/Collaboration/SARP phantom]] (Keith Gunderson)<br />
#[[FMA (Protege) links to Slicer]] (Vish, Mike, Florin, Jim, Steve, Wendy)<br />
#[[NA-MIC/Projects/External Collaboration/Measuring Alcohol and Stress Interaction]]<br />
#[[NA-MIC/Projects/External Collaboration/Slicer3-vmtk Integration]] (Luca Antiga, Dan Blezek, Mike Halle, Steve Pieper)<br />
#[[NA-MIC/Projects/External Collaboration/Mesh Generation Summer 2008]] (Iowa Group)<br />
#[[NA-MIC/Projects/Collaboration/Carto_scar_BIDMC]] (Dana Peters Group)<br />
#[[NA-MIC/Projects/Collaboration/3D Ultrasound Module in Slicer3]] (Junichi, Haiying and Noby - SPL, David - Queen's), Danielle - Robarts) )<br />
#[[NA-MIC/Projects/Collaboration/MGH RadOnc]] (Greg Sharp, MGH)<br />
#[[NA-MIC/Projects/External Collaboration/W&M CRTC]] Non-rigid registration for neurosurgery (Nikos Chrisochoides, Andriy Fedorov, College of William&Mary)<br />
#[[NA-MIC/Projects/Collaboration/SBIA UPenn]] Non-rigid DTI Registration(Ragini Verma, SBIA Upenn)<br />
<br />
===Non-Medical Collaborations===<br />
#[[NA-MIC/Projects/Non-Medical Collaborations/Astronomical Medicine|Astronomical Medicine]] (Harvard IIC: Douglas Alan, Michael Halle)<br />
<br />
== Preparation ==<br />
<br />
# Please make sure that you are on the http://public.kitware.com/cgi-bin/mailman/listinfo/na-mic-project-week mailing list<br />
<br />
# [[Engineering:TCON_2008|May 08 and May 15 TCON DBPs ONLY]] at 3pm ET to discuss NA-MIC DBP Projects ONLY. <br />
# [[Engineering:TCON_2008|May 22 TCON#1]] at 3pm ET to discuss NA-MIC Engr Core Projects and Assign/Verify Teams<br />
# [[Engineering:TCON_2008|May 29 TCON#2]] at 3pm ET to discuss NA-MIC ALGORITHMS Core Lead Projects. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 5 TCON#3]] at 3pm ET to discuss NA-MIC EXTERNAL Collaborations. All NIH funded "collaborations with NCBC" leads should call. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 12 TCON#4]] at 3pm ET to discuss NA-MIC EXTERNAL Collaborations. All other collaboration leads should call. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 19 TCON#5]] at 3pm ET to tie loose ends. Anyone with un-addressed questions should call.<br />
# By 3pm ET on June 12, 2008: [[NA-MIC/Projects/Theme/Template|Complete a templated wiki page for your project]]. Please do not edit the template page itself, but create a new page for your project and cut-and-paste the text from this template page. If you have questions, please send an email to tkapur at bwh.harvard.edu.<br />
# By 3pm on June 19, 2008: Create a directory for each project on the [[Engineering:SandBox|NAMIC Sandbox]] (Zack)<br />
## Commit on each sandbox directory the code examples/snippets that represent our first guesses of appropriate methods. (Luis and Steve will help with this, as needed)<br />
## Gather test images in any of the Data sharing resources we have (e.g. the BIRN). These ones don't have to be many. At least three different cases, so we can get an idea of the modality-specific characteristics of these images. Put the IDs of these data sets on the wiki page. (the participants must do this.)<br />
## Setup nightly tests on a separate Dashboard, where we will run the methods that we are experimenting with. The test should post result images and computation time. (Zack)<br />
# Please note that by the time we get to the project event, we should be trying to close off a project milestone rather than starting to work on one...<br />
<br />
==Attendee List==<br />
# Jack Blevins Acoustic Med<br />
# Pratik Patel Brainlab<br />
# Mark Anderson BWH<br />
# Nicole Aucoin BWH<br />
# Sylvain Bouix BWH<br />
# Michael Halle BWH<br />
# Nobuhiko Hata BWH<br />
# Katie Hayes BWH<br />
# Scott Hoge BWH<br />
# Marianna Jakab BWH<br />
# Tina Kapur BWH<br />
# Ron Kikinis BWH<br />
# Jacek Kukluk BWH<br />
# Haying Liu BWH<br />
# Bjoern Menze BWH<br />
# Wendy Plesniak BWH<br />
# Kilian Pohl BWH<br />
# Sonia Pujol BWH<br />
# Carlos Sánchez Mendoza BWH<br />
# Priya Srinivasan BWH<br />
# Junichi Tokuda BWH<br />
# Demian Wasserman BWH (INRIA)<br />
# C-F Westin BWH<br />
# Xiaodong Tao GE<br />
# Dirk Padfield GE<br />
# Jim Miller GE<br />
# Surprise Guest from EAB<br />
# Viswanath Avasarala GE<br />
# John Melonakos GA Tech<br />
# Yi Gao GA Tech<br />
# Tauseef Rehman GA Tech<br />
# Sean Megason Harvard Med<br />
# Alex Gouaillard Harvard Med<br />
# Kishore Mosaliganti Harvard Med<br />
# Arnaud Gelas Harvard Med<br />
# Dana Peters Harvard Med<br />
# Jason Taclas Harvard Med<br />
# Douglas Alan Harvard<br />
# Toru Higaki Hiroshima U<br />
# Daniel Blezek Isomics<br />
# Curtis Lisle Isomics<br />
# Steve Pieper Isomics<br />
# Alex Yarmarkovich Isomics<br />
# Csaba Csoma JHU<br />
# Peter Kazanzides JHU<br />
# Will Schroeder Kitware<br />
# Sebastien Barre Kitware<br />
# Julien Jomier Kitware<br />
# Bill Hoffman Kitware<br />
# Luis Ibanez Kitware<br />
# Luca Antiga Mario Negri Inst<br />
# Randy Gollub MGH<br />
# Silas Mann MGH<br />
# Greg Sharp MGH<br />
# Marta Peroni MGH<br />
# Serdar Balci MIT<br />
# Bryce Kim MIT<br />
# Clare Poynton MIT<br />
# Tammy Riklin Raviv MIT<br />
# Polina Golland MIT<br />
# Jeremy Bockholt MRN Lupus DBP<br />
# Mark Scully MRN Lupus DBP<br />
# Gabor Fichtinger Queen's<br />
# David Gobbi Queen's<br />
# Purang Abolmaesumi Queen's<br />
# Siddharth Vikal Queen's<br />
# Zhen Qian Rutgers<br />
# Jinghao Zhou Rutgers<br />
# Jeffrey Grethe UCSD<br />
# Marco Ruiz UCSD<br />
# Chris Churas UCSD<br />
# Nathan Hageman UCLA<br />
# Keith Gunderson U Iowa<br />
# Gary Christensen U Iowa<br />
# Jeffrey Hawley U Iowa<br />
# Kate Raising U Iowa<br />
# Nathan Fritze U Iowa<br />
# Paul Song U Iowa<br />
# Cheng Zhang U Iowa<br />
# Ying Wei U Iowa<br />
# Nathan Burnette U Iowa<br />
# Hans Johnson U Iowa<br />
# Vincent Magnotta U Iowa<br />
# Clement Vachet UNC<br />
# Zhexing Liu UNC<br />
# Ragini Verma U Penn<br />
# Luke Bloy U Penn<br />
# Yang Li U Penn<br />
# Ran Tao Utah<br />
# Marcel Prastawa Utah<br />
# Casey Goodlett Utah<br />
# Ross Whitaker Utah<br />
# John Hale U Tulsa<br />
# Cody Pollet U Tulsa<br />
# Nikeisha Schimke U Tulsa<br />
# Adam Wittek Western Australia<br />
# Grand Joldes Western Australia<br />
# Jamie Berger Western Australia<br />
# Carling Cheung Western Ontario<br />
# Danielle Pace Western Ontario<br />
# Vidya Rajagopalan VA Tech<br />
# Nikos Chrisochoides William and Mary<br />
# Andriy Fedorov William and Mary<br />
# Dan Marcus Washington U<br />
# Tim Olsen Washington U<br />
# Kevin Archie Washington U<br />
# Misha Milchenko Washington U<br />
# Xenophon Papademetris Yale U<br />
# John Onofrey Yale U<br />
# Yifeng Jiang Yale U<br />
# Dustin Scheinost Yale U<br />
#* Please note that Registration is closed. Do not add names here. [[User:Tkapur|Tkapur]] 18:19, 19 June 2008 (UTC)<br />
<br />
==Pictures==<br />
To be added.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24724Projects:GroupwiseRegistration2008-05-16T20:51:39Z<p>Serdar: </p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 5: Central slices of 3D volumes for groupwise registration. Rows show mean and standard deviation images followed by label overlap images for GM, WM and CSF labels. Columns display the results for affine and B-splines with grid spacing 32, 16 and 8 voxels, respectively. ]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 6: GM, WM DICE measures computed for different deformation field resolution levels. Blue bars show the results for groupwise registration and the red bars show the results for registration to the mean setting.]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 7: DICE measures for manually segmented labels. Bars correspond to the same setting as in figure 6. ]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=+balci+AND+groupwise+AND+registration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| link to papers]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT&diff=24722Algorithm:MIT2008-05-16T20:50:00Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
__NOTOC__<br />
= Overview of MIT Algorithms =<br />
<br />
Our group seeks to model statistical variability of anatomy and function across subjects and between populations and to utilize computational models of such variability to improve predictions for individual subjects, as well as characterize populations. Our long-term goal is to develop methods for joint modeling of anatomy and function and to apply them in clinical and scientific studies. We work primarily with anatomical, DTI and fMRI images. We actively contribute implementations of our algorithms to the NAMIC-kit.<br />
<br />
= MIT Projects =<br />
<br />
{| cellpadding="10"<br />
| style="width:15%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|200px]]<br />
| style="width:85%" | <br />
<br />
== [[Projects:ShapeBasedSegmentationAndRegistration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithm assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Projects:ShapeBasedSegmentationAndRegistration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Medical Image Analysis,11(6), pp. 465-477, 2007. <b>Best Paper Award MICCAI 2006 </b><br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|200px]]<br />
| |<br />
<br />
== [[Projects:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Projects:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|200px]]<br />
| |<br />
<br />
== [[Projects:MultimodalAtlas|Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Projects:MultimodalAtlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|200px]]<br />
| |<br />
<br />
== [[Projects:GroupwiseRegistration|Groupwise Registration]] ==<br />
<br />
We extend a previously demonstrated entropy based groupwise registration method to include a free-form deformation model based on B-splines. We provide an efficient implementation using stochastic gradient descents in a multi-resolution setting. We demonstrate the method in application to a set of 50 MRI brain scans and compare the results to a pairwise approach using segmentation labels to evaluate the quality of alignment.<br />
[[Projects:GroupwiseRegistration|More...]]<br />
<br />
S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|200px|]]<br />
| |<br />
<br />
== [[Projects:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Projects:ShapeAnalysisWithOvercompleteWavelets|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007.<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|200px]]<br />
| |<br />
<br />
== [[Projects:fMRIClustering|fMRI clustering]] ==<br />
<br />
In this project we study the application of model-based clustering algorithms in identification of functional connectivity in the brain. [[Projects:fMRIClustering|More...]]<br />
<br />
<font color="red">'''New: '''</font> P. Golland, Y. Golland, R. Malach. Detection of Spatial Activation Patterns As Unsupervised Segmentation of fMRI Data. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 110-118, 2007. <br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|200px]]<br />
| |<br />
<br />
== [[Projects:DTIFiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Projects:DTIFiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|200px]]<br />
| |<br />
<br />
<br />
<br />
== [[Projects:DTIClustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Projects:DTIClustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby. Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors. Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|200px]]<br />
| |<br />
<br />
== [[Projects:DTIModeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Projects:DTIModeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. E. L. Grimson, S. K. Warfield, W. M. Wells, A Unified Framework for Clustering and Quantitative Analysis of White Matter Fiber Tracts. Medical Image Analysis, in press. <br />
<br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts, IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|200px]]<br />
| |<br />
<br />
== [[Projects:fMRIDetection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Projects:fMRIDetection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|200px]]<br />
| |<br />
<br />
== [[Projects:DTISegmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Projects:DTISegmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|200px]]<br />
| |<br />
<br />
== [[Projects:DTIStochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Projects:DTIStochasticTractography|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|200px]]<br />
| |<br />
<br />
== [[Projects:ShapeAnalysis|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Projects:ShapeAnalysis|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.<br />
<br />
<br />
== [[Projects:HippocampalSubfieldSegmentation|Model-Based Segmentation of Hippocampal Subfields in In Vivo MRI]] ==<br />
<br />
<br />
|}</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24684Projects:GroupwiseRegistration2008-05-16T20:12:06Z<p>Serdar: </p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 5: Central slices of 3D volumes for groupwise registration. Rows show mean and standard deviation images followed by label overlap images for GM, WM and CSF labels. Columns display the results for affine and B-splines with grid spacing 32, 16 and 8 voxels, respectively. ]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 6: GM, WM DICE measures computed for different deformation field resolution levels. Blue bars show the results for groupwise registration and the red bars show the results for registration to the mean setting.]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 7: DICE measures for manually segmented labels. Bars correspond to the same setting as in figure 6. ]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=+balci+AND+groupwise+AND+registration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| links to papers]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24674Projects:GroupwiseRegistration2008-05-16T19:59:58Z<p>Serdar: </p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 5: Central slices of 3D volumes for groupwise registration. Rows show mean and standard deviation images followed by label overlap images for GM, WM and CSF labels. Columns display the results for affine and B-splines with grid spacing 32, 16 and 8 voxels, respectively. ]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 6: GM, WM DICE measures computed for different deformation field resolution levels. Blue bars show the results for groupwise registration and the red bars show the results for registration to the mean setting.]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 7: DICE measures for manually segmented labels. Bars correspond to the same setting as in figure 6. ]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
''In Print''<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=balci&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| NA-MIC Publications Database]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24672Projects:GroupwiseRegistration2008-05-16T19:51:11Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 5: Central slices of 3D volumes for groupwise registration. Rows show mean and standard deviation images followed by label overlap images for GM, WM and CSF labels. Columns display the results for affine and B-splines with grid spacing 32, 16 and 8 voxels, respectively. ]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 6: GM, WM DICE measures computed for different deformation field resolution levels. Blue bars show the results for groupwise registration and the red bars show the results for registration to the mean setting.]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 7: DICE measures for manually segmented labels. Bars correspond to the same setting as in figure 6. ]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
''In Print''<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| NA-MIC Publications Database]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24671Projects:GroupwiseRegistration2008-05-16T19:47:01Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 5: Central slices of 3D volumes for groupwise registration. Rows show mean and standard deviation images followed by label overlap images for GM, WM and CSF labels. Columns display the results for affine and B-splines with grid spacing 32, 16 and 8 voxels, respectively. ]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
''In Print''<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| NA-MIC Publications Database]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24670Projects:GroupwiseRegistration2008-05-16T19:42:37Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
''In Print''<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| NA-MIC Publications Database]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=24669Projects:GroupwiseRegistration2008-05-16T19:41:56Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Internal_Collaborations:StructuralImageAnalysis|NA-MIC Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
__NOTOC__<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure 5 look visually similar. From Figures 6 and 7, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure 5 show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average over the whole population.<br />
<br />
= Key Investigators =<br />
<br />
* MIT Algorithms: S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
''In Print''<br />
<br />
*[http://www.na-mic.org/pages/Special:Publications?text=Projects%3AGroupwiseRegistration&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked| NA-MIC Publications Database]<br />
<br />
[[Category: Registration]] [[Category:Segmentation]] [[Category:MRI]] [[Category:Schizophrenia]]</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Summer_Project_Week&diff=245082008 Summer Project Week2008-05-09T13:32:40Z<p>Serdar: /* Attendee List */</p>
<hr />
<div>Back to [[Engineering:Programming_Events|Programming/Project Events]]<br />
<br />
[[Image:ProjectWeek-2008.png|thumb|220px|right|Summer 2008]]<br />
<br />
== Logistics ==<br />
<br />
'''Dates:''' June 23-27, 2008<br />
<br />
'''Location:''' MIT. [[Meeting_Locations:MIT_Grier_A_%26B|Grier Rooms A & B: 34-401A & 34-401B]].<br />
<br />
<br />
'''Registration Fee:''' $260 (this will cover the cost of breakfast, lunch and coffee breaks for the week). Due by Friday, June 13th, 2008. Please make checks out to "Massachusetts Institute of Technology" and mail to: Donna Kaufman, MIT, 77 Massachusetts Ave., 38-409b, Cambridge, MA 02139<br />
<br />
If you are attending for one day only, the registration fee is not required.<br />
<br />
'''Hotel:''' We have a group rate of TBD/night at the [http://www.hotelatmit.com Hotel at MIT]. (Use group code NAM.) Here is some information about several other Boston area hotels that are convenient to NA-MIC events: [[Boston_Hotels|Boston_Hotels]]. Summer is tourist season in Boston, so please book your rooms early.<br />
<br />
([[Project Week Logistics Checklist|This is a checklist for the onsite planning items]])<br />
<br />
==Introduction to NA-MIC Project Week==<br />
<br />
NA-MIC Project Week is a hands on activity -- programming using the [[NA-MIC-Kit|NA-MIC Kit]], algorithm design, and clinical application -- that has become one of the major events in the [[NA-MIC-Kit|NA-MIC Kit]] calendar. This event is the seventh of the [[Engineering:Programming_Events|'''series''']]. It is held in the summer at MIT (typically the last week of June), and a shorter version is held in Salt Lake City in the winter (typically the second week of January). <br />
The main goal of these events if to move forward the deliverables of NA-MIC. NA-MIC participants and their collaborators are welcome to attend. <br />
<br />
* NA-MIC Members: Participation in this event is voluntary -- if you don't think this will help you move forward in your work, there is no obligation to attend.<br />
* Ideal candidates are those who want to contribute to the [[NA-MIC-Kit|NA-MIC Kit]], and those who can help make it happen.<br />
* This is not an introduction to the components of the [[NA-MIC-Kit|NA-MIC Kit]].<br />
* NA-MIC Core 1 (Algorithms) - bring your algorithms and code to work on in the company of Core 2 engineers and Core 3 scientists.<br />
* NA-MIC Core 2 (Engineering) - bring your code for infrastructure and applications to extend the [[NA-MIC-Kit|NA-MIC Kit]] capabilities, integrate Core 1 algorithms, and refine workflows for Core 3.<br />
* NA-MIC Core 3 (DBP) - bring your data to work on with the [[NA-MIC-Kit|NA-MIC Kit]] and get assistance and provide feedback to Core 1 scientists and Core 2 engineers.<br />
* External Collaborators - if you are working on a project that uses the [[NA-MIC-Kit|NA-MIC kit]], and want to participate to get help from NA-MIC Engineering, please send an email to Tina Kapur (tkapur at bwh.harvard.edu). Please note that the event is open to people outside NA-MIC, subject to availability.<br />
* Everyone should '''bring a laptop'''. We will have four projectors.<br />
* About half the time will be spent working on projects and the other half in project related discussions.<br />
* You '''do''' need to be actively working on a NA-MIC related project in order to make this investment worthwhile for everyone.<br />
<br />
=== Agenda===<br />
* Monday <br />
** noon-1pm lunch <br />
**1pm: Welcome (Ron Kikinis)<br />
** 1:05-3:30pm Introduce [[#Projects|Projects]] using templated wiki pages (all Project Leads) ([[NA-MIC/Projects/Theme/Template|Wiki Template]]) <br />
** 3:30-5:30pm Start project work<br />
* Tuesday <br />
** 8:30am breakfast<br />
** 9:00-9:45am: NA-MIC Software Process <br />
** 10-10:30am [[Project Week 2008 Slicer 3.0 Update|Slicer 3.0 Update]] (Jim Miller, Steve Pieper)<br />
** noon lunch<br />
** 2:30-3:30pm: [[Project Week 2008 Special topic breakout: Non-Linear Registration]] <br />
** 5:30pm adjourn for day<br />
* Wednesday <br />
** 8:30am breakfast<br />
** 9:00-12pm [[Project Week 2008 Special topic breakout: ITK]] (Luis Ibanez)<br />
** noon lunch<br />
** 2:30-3:30pm: [[Project Week 2008 Special topic breakout: XNAT Database]] (Daniel Marcus)<br />
** 5:30pm adjourn for day<br />
* Thursday<br />
** 8:30am breakfast<br />
** noon lunch<br />
** 5:30pm adjourn for day<br />
* Friday <br />
** 8:30am breakfast<br />
** 10am-noon: Project Progress using update [[#Projects|Project Wiki pages]]<br />
** noon lunch boxes and adjourn. (Next one [[AHM_2009| in Utah the week of Jan 5, 2009]])<br />
<br />
=== Preparation ===<br />
<br />
# Please make sure that you are on the http://public.kitware.com/cgi-bin/mailman/listinfo/na-mic-project-week mailing list<br />
<br />
# [[Engineering:TCON_2008|May 08 and May 15 TCON DBPs ONLY]] at 3pm ET to discuss NA-MIC DBP Projects ONLY. <br />
# [[Engineering:TCON_2008|May 22 TCON#1]] at 3pm ET to discuss NA-MIC Engr Core Projects and Assign/Verify Teams<br />
# [[Engineering:TCON_2008|May 29 TCON#2]] at 3pm ET to discuss NA-MIC ALGORITHMS Core Lead Projects. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 5 TCON#3]] at 3pm ET to discuss NA-MIC EXTERNAL Collaborations. All NIH funded "collaborations with NCBC" leads should call. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 12 TCON#4]] at 3pm ET to discuss NA-MIC EXTERNAL Collaborations. All other collaboration leads should call. Project leads should sign up for a slot [[Engineering:TCON_2008|here]]. Projects will be discussed in order of the signups. <br />
# [[Engineering:TCON_2008|June 12 TCON#4]] at 3pm ET to tie loose ends. Anyone with un-addressed questions should call.<br />
# By 3pm ET on June 12, 2008: [[NA-MIC/Projects/Theme/Template|Complete a templated wiki page for your project]]. Please do not edit the template page itself, but create a new page for your project and cut-and-paste the text from this template page. If you have questions, please send an email to tkapur at bwh.harvard.edu.<br />
# By 3pm on June 19, 2008: Create a directory for each project on the [[Engineering:SandBox|NAMIC Sandbox]] (Zack)<br />
## Commit on each sandbox directory the code examples/snippets that represent our first guesses of appropriate methods. (Luis and Steve will help with this, as needed)<br />
## Gather test images in any of the Data sharing resources we have (e.g. the BIRN). These ones don't have to be many. At least three different cases, so we can get an idea of the modality-specific characteristics of these images. Put the IDs of these data sets on the wiki page. (the participants must do this.)<br />
## Setup nightly tests on a separate Dashboard, where we will run the methods that we are experimenting with. The test should post result images and computation time. (Zack)<br />
# Please note that by the time we get to the project event, we should be trying to close off a project milestone rather than starting to work on one...<br />
<br />
== A History in Wiki Links ==<br />
<br />
A history of all the programming/project events in NA-MIC is available by following [[Engineering:Programming_Events|this link]].<br />
<br />
== Projects ==<br />
<br />
===DBP II===<br />
These are projects by the new set of DBPS:<br />
#[[DBP2:Harvard|Velocardio Facial Syndrome (VCFS) as a Genetic Model for Schizophrenia]] (Harvard: Marek Kubicki, PI)<br />
#[[DBP2:UNC|Longitudinal MRI Study of Early Brain Development in Autism]] (UNC: Heather Hazlett, Joseph Piven, PI)<br />
#[[DBP2:MIND|Analysis of Brain Lesions in Lupus]] (MIND/UNM: Jeremy Bockholt, Charles Gasparovic PI)<br />
#[[DBP2:JHU|Segmentation and Registration Tools for Robotic Prostate Intervention]] (Queens/JHU: Gabor Fichtinger, PI)<br />
<br />
===Structural Analysis===<br />
<br />
===Diffusion Image Analysis===<br />
<br />
===Calibration/Validation===<br />
This is a new category of projects jointly led by team members in Core 1, Core 3 and Core 5<br />
# [[Projects/Diffusion/2007_Project_Week_Contrasting_Tractography_Measures]] (Westin, Gollub, Gerig, Whitaker, Pujol)<br />
<br />
===NA-MIC Kit - Slicer 3===<br />
<br />
===External Collaborations===<br />
<br />
===Non-Medical Collaborations===<br />
<br />
==Attendee List==<br />
# Ron Kikinis, SPL<br />
# Gary Christensen, The University of Iowa<br />
# Jeffrey Hawley, Gary Christensen's student<br />
# Kate Raising, Gary Christensen's student<br />
# Nathan Fritze, Gary Christensen's student<br />
# Paul Song, Gary Christensen's student<br />
# Cheng Zhang, Gary Christensen's student<br />
# Ying Wei, Gary Christensen's student<br />
# Nathan Burnette, The University of Iowa<br />
# Steve Pieper, Isomics, Core 2/6<br />
# Dana C. Peters, BIDMC Harvard Medical<br />
# Jason Taclas, Student, BIDMC Harvard Medical<br />
# Nicole Aucoin, BWH, Core 2<br />
# Will Schroeder, Kitware, Cores 2/4<br />
# Sebastien Barre, Kitware, Core 2<br />
# Julien Jomier, Kitware, Core 2<br />
# Curtis Lisle, KnowledgeVis, Core 2<br />
# Katie Hayes, BWH, Core 2<br />
# Randy Gollub, MGH, Core 5<br />
# Clement Vachet, UNC, Core 3<br />
# Tauseef Rehman, GA Tech, Core 1<br />
# Jeffrey Grethe, UCSD, Core 2<br />
# Marco Ruiz, UCSD, Core 2<br />
# Zhen Qian, Rutgers University<br />
# Jinghao Zhou, Rutgers University<br />
# Luca Antiga, Mario Negri Institute<br />
# Adam Wittek, The University of Western Australia<br />
# Grand Joldes, The University of Western Australia<br />
# Serdar Balci, MIT<br />
<br />
==Pictures==</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Core_1_Core_3_mtg&diff=236842008 Core 1 Core 3 mtg2008-04-08T02:05:47Z<p>Serdar: </p>
<hr />
<div>* Dates: May 22-23. Starts morning of May 22 and finishes afternoon of May 23.<br />
* Location: Warnock Engineering Building, University of Utah ([http://www.sci.utah.edu/map.html Directions]).<br />
<br />
* Logistics: TBA<br />
<br />
* Attendees:<br />
** Ross Whitaker<br />
** Tom Fletcher<br />
** Guido Gerig<br />
** Martin Styner<br />
** John Melonakos<br />
** Yi Gao<br />
** H Jeremy Bockholt<br />
** Sylvain Bouix<br />
** Serdar Balci</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_BSplineReg&diff=211952008 Winter Project Week BSplineReg2008-01-11T15:18:54Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| <br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
<br />
We run an experiment on a machine<br />
with 8GB of memory, 4 core CPU <br />
running gcc compiler on a linux system. <br />
<br />
We used a B-Spline with mesh size 32x32x16<br />
with an image of size 256x256x124 and used<br />
Mattes mutual information as the registration algorithm.<br />
<br />
Run-times:<br />
* 21 seconds with caching (2.4 GB of memory)<br />
* 52 seconds w/o caching (800MB of memory)<br />
<br />
Overall in this experiment, caching improved the performance by 2.5 while using 3 times more memory.<br />
We are in the process of exploring the effect of number of samples, image size and compiler on <br />
registration time.<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div></div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211942008 Winter Project Week GroupwiseReg2008-01-11T15:17:55Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Brad Davis<br />
* SCI: Casey Goodlett<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to expand the NAMIC Software Kit with the addition of an image-based atlas<br />
building algorithm that was developed by Polina Golland’s lab at MIT. The basic functionality of this software will<br />
be demonstrated to the NAMIC community at the January 2008 NAMIC all-hands meeting. A secondary<br />
goal is to begin the process of gathering usability and interface requests from targeted clinical collaborators.<br />
It is intended that users’ requests will be incorporated in to a future project aimed at tailoring this software<br />
for widespread NAMIC use.<br />
<br />
See our [[Projects:GroupwiseRegistration| Algorithm Project Page]] for more details.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
The registration algorithms are already available as C++ sourcecode. We plan to integrate them into the NAMIC community by:<br />
<br />
*Adapt sourcecode to utilize the command-line parsing mechanism that was designed for Slicer3 external modules. <br />
*Create a simple Slicer3 graphical user interface (external module) that allows a user to generate an atlas from their data. <br />
*User Feedback. During the January 2008 NAMIC all-hands meeting we will solicit feedback on this project from potential collaborators.<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_BSplineReg&diff=211932008 Winter Project Week BSplineReg2008-01-11T15:16:45Z<p>Serdar: New page: {| |Return to [[2008_Winter_Project_Week ]] |valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of ima...</p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
<br />
We run an experiment on a machine<br />
with 8GB of memory, 4 core CPU <br />
running gcc compiler on a linux system. <br />
<br />
We used a B-Spline with mesh size 32x32x16<br />
with an image of size 256x256x124 and used<br />
Mattes mutual information as the registration algorithm.<br />
<br />
Run-times:<br />
* 21 seconds with caching (2.4 GB of memory)<br />
* 52 seconds w/o caching (800MB of memory)<br />
<br />
Overall in this experiment, caching improved the performance by 2.5 while using 3 times more memory.<br />
We are in the process of exploring the effect of number of samples, image size and compiler on <br />
registration time.<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div></div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211922008 Winter Project Week GroupwiseReg2008-01-11T15:11:56Z<p>Serdar: /* References */</p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
<br />
We run an experiment on a machine<br />
with 8GB of memory, 4 core CPU <br />
running gcc compiler on a linux system. <br />
<br />
We used a B-Spline with mesh size 32x32x16<br />
with an image of size 256x256x124 and used<br />
Mattes mutual information as the registration algorithm.<br />
<br />
Run-times:<br />
* 21 seconds with caching (2.4 GB of memory)<br />
* 52 seconds w/o caching (800MB of memory)<br />
<br />
Overall in this experiment, caching improved the performance by 2.5 while using 3 times more memory.<br />
We are in the process of exploring the effect of number of samples, image size and compiler on <br />
registration time.<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div></div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211912008 Winter Project Week GroupwiseReg2008-01-11T15:11:31Z<p>Serdar: /* Jan 2008 Project Week */</p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
<br />
We run an experiment on a machine<br />
with 8GB of memory, 4 core CPU <br />
running gcc compiler on a linux system. <br />
<br />
We used a B-Spline with mesh size 32x32x16<br />
with an image of size 256x256x124 and used<br />
Mattes mutual information as the registration algorithm.<br />
<br />
Run-times:<br />
* 21 seconds with caching (2.4 GB of memory)<br />
* 52 seconds w/o caching (800MB of memory)<br />
<br />
Overall in this experiment, caching improved the performance by 2.5 while using 3 times more memory.<br />
We are in the process of exploring the effect of number of samples, image size and compiler on <br />
registration time.<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211902008 Winter Project Week GroupwiseReg2008-01-11T15:11:06Z<p>Serdar: /* Jan 2008 Project Week */</p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
We run an experiment on a machine<br />
with 8GB of memory, 4 core CPU <br />
running gcc compiler on a linux system. <br />
<br />
We used a B-Spline with mesh size 32x32x16<br />
with an image of size 256x256x124 and used<br />
Mattes mutual information as the registration algorithm.<br />
<br />
Run-times:<br />
* 21 seconds with caching (2.4 GB of memory)<br />
* 52 seconds w/o caching (800MB of memory)<br />
<br />
Overall in this experiment, caching improved the performance by 2.5 while using 3 times more memory.<br />
We are in the process of exploring the effect of number of samples, image size and compiler on <br />
registration time.<br />
<br />
<br />
</div><br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211892008 Winter Project Week GroupwiseReg2008-01-11T15:04:40Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to develop an efficient B-Splines based registration code.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211882008 Winter Project Week GroupwiseReg2008-01-11T15:03:41Z<p>Serdar: /* Key Investigators */</p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Luis Ibanez<br />
<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to come up with an efficient registration code using B-Splines as a deformation model.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_GroupwiseReg&diff=211872008 Winter Project Week GroupwiseReg2008-01-11T15:03:00Z<p>Serdar: </p>
<hr />
<div>{|<br />
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]<br />
|valign="top"| [[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
|}<br />
<br />
__NOTOC__<br />
===Key Investigators===<br />
* MIT: Serdar Balci, Polina Golland<br />
* Kitware: Brad Davis<br />
* SCI: Casey Goodlett<br />
<br />
<div style="margin: 20px;"><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Objective</h1><br />
<br />
The goal of this project is to come up with an efficient registration code using B-Splines as a deformation model.<br />
We explore the space of different optimization schemes e.g. caching, sparsity and memory-speed trade-off's.<br />
<br />
</div><br />
<br />
<div style="width: 27%; float: left; padding-right: 3%;"><br />
<br />
<h1>Approach, Plan </h1><br />
<br />
We run benchmark experiments in the special case of itkMattesMutualInformation implementation. <br />
The current implementation in ITK uses caching of B-Spline coefficient in order to speed-up <br />
the registration. Also the sparsity of B-Spline coefficient updates are taken into account, yielding <br />
and efficient implementation.<br />
<br />
The drawback of caching is the use of high amounts of memory. <br />
We compare the performance of caching vs. not-caching(computing the coefficients on the fly)<br />
by deactivating caching code in itkMattesMutualInformation .<br />
<br />
</div><br />
<br />
<div style="width: 40%; float: left;"><br />
<br />
<h1>Progress</h1><br />
<br />
<br />
====Jan 2008 Project Week====<br />
<br />
</div><br />
<br />
<br style="clear: both;" /><br />
<br />
</div><br />
<br />
<br />
===References===<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week&diff=211862008 Winter Project Week2008-01-11T14:50:28Z<p>Serdar: /* Other NA-MIC Projects */</p>
<hr />
<div>Back to [[Project Events]], [[AHM_2008]], [[Events]]<br />
__NOTOC__<br />
== Projects ==<br />
*2008 Project Week [[2008_Winter_Project_Week_Template|'''Template''']]<br />
*[[2007_Programming/Project_Week_MIT#Projects|Last Year's Projects as a reference]]<br />
*For hosting projects, we are planning to make use of the NITRC resources. See [[NA-MIC_and_NITRC | Information about NITRC Collaboration]]<br />
*Next Project Week is at MIT -- June 23-27, 2008<br />
<br />
<br />
===Introduction to NA-MIC Project Week===<br />
<br />
Please read an introduction about these events [[Project_Events#Introduction|here]].<br />
<br />
===NA-MIC DBP Roadmap Projects===<br />
Please note that these projects correspond to four clinical Roadmap application projects that will be pursued in focused parallel tracks at the meeting, each corresponding to a DBP problem. <br />
<br />
#[[2008_Winter_Project_Week:StochasticTract_Arcuate|Stochastic tractography of the arcuate fasciculus in schizophrenia]] (Marek Kubicki, Tri Ngo, Doug Markant) [[DBP2:Harvard:Brain_Segmentation_Roadmap|[Harvard Roadmap Project: Stochastic Tractography for VCFS]]]<br />
#[[2008_Winter_Project_Week:Cortical_Thickness|Cortical thickness analysis of pediatric brains]] (Heather Cody Hazlett, Martin Styner, Clement Vachet, Jim Miller) [[DBP2:UNC:Cortical_Thickness_Roadmap|[UNC Roadmap Project: Cortical Thickness Measurement for Autism]]]<br />
#[[2008_Winter_Project_Week:Lesions|Towards an end to end lesion analysis feature in Slicer3]] (Mark Scully, Jeremy Bockholt, Brad Davis, Marcel Prastawa, Sonia Pujol, Vincent Magnotta)[[DBP2:MIND:Roadmap|[MIND Roadmap Project: Brain Lesion Analysis in Lupus]]]<br />
#[[2008_Winter_Project_Week:Robotic_Prostate_Interventions |Robotic Prostate Interventions]] (David Gobbi, Csaba Csoma, Junichi Tokuda, Katie Hayes)[[DBP2:JHU:Roadmap|[JHU Roadmap Project: Segmentation and Registration for Robotic Prostate Intervention]]]<br />
#[[2008_Winter_Project_Week:Prostate_Segmentation|MRI-US Prostate Segmentation]], (Yi Gao, Ponnappan Arumuganainar, John Melonakos, Allen Tannenbaum, Gabor Fichtinger, Xiaodong Tao)<br />
<br />
===Other NA-MIC Projects===<br />
# [[2008_Winter_Project_Week_GroupwiseReg | Groupwise Registration and Atlas Building]] (Brad Davis, Serdar Balci, Casey Goodlett)<br />
# [[2008_Winter_Project_Week_BSplineReg | Optimization for B-Spline based Registration]] (Serdar Balci, Luis Ibanez, Polina Golland)<br />
# [[2008_Winter_Project_Week:MRISC|Joint Segmentation and Classification of MR Images Based on Structure-specific Affine Registration]] (Mert Sabuncu, Kilian Pohl, Xiaodong Tao)<br />
<!-- # [[2008_Winter_Project_Week:CorPar|Cortical Surface Parcellation]] (Thomas Yeo, Mert Sabuncu,Luis Ibanez,Brad Davis)--><br />
# [[2008_Winter_Project_Week:Dorsolateral_Prefrontal_Cortex_Segmentation|Dorsolateral Prefrontal Cortex Segmentation]] (Marek Kubicki, Sylvain Bouix, John Melonakos, Brad Davis, Polina Golland)<br />
#[[2008_Winter_Project_Week:Particle_Correspondence_DTI|Incorporating DTI data into entropy-based particle system for cortical correspondence]] (Ipek Oguz, Josh Cates, Tom Fletcher, Martin Styner, Xiaodong Tao)<br />
#[[2008_Winter_Project_Week:Population_DTI|Integrating population based DTI tools into NAMIC Kit]] (Casey Goodlett, Alex Yarmakovich)<br />
#[[2008_Winter_Project_Week:Population_DTI_Application|Application of population based DTI tools to Schizophrenia]] (Casey Goodlett, Alex Yarmakovich, Marek Kubicki)<br />
#[[2008_Winter_Project_Week:Geodesic_Tractography_Segmentation|Geodesic Tractography Segmentation]], (John Melonakos, Luis Ibanez, Marek Kubicki)<br />
#[[2008_Winter_Project_Week:Fluid_Mechanics_Tractography|Fluid Mechanics Based DTI Tractography]] (Nathan Hageman,Alex Yarmakovich, Jim Miller)<br />
#[[2008_Winter_Project_Week:MRMLScenesForExecutionModel |MRML Scenes for the Execution Model including Transforms]] (Jim Miller, Brad Davis, Nicole Aucoin, Alex Yarmarkovich, Steve Pieper)<br />
#[[2008_Winter_Project_Week:UnstructuredGrids | Unstructured Grids and Mesh Support]] (Curt, Alex, Steve, Will, Vince, Bob O'Bara)<br />
#[[2008_Winter_Project_Week:PythonSupport |Python Support in Slicer 3]] (Luca Antiga, Dan Blezek)<br />
#[[2008_Winter_Project_Week:MRMLTransformHardening |Transform hardening in MRML]] (Luca Antiga, Steve Pieper)<br />
#[[2008_Winter_Project_Week:OutOfCoreMRML | MRML support for out of core processing with fMRI and DTI as use cases]] (Steve Pieper, Jim Miller, Wendy Plesniak, Alex Yarmakovich, Will Schroeder)<br />
#[[2008_Winter_Project_Week:EventBrokerInSlicer3 |Event broker in Slicer3]] (Jim Miller, Steve Pieper, Alex Yarmakovich, Luca Antiga, Dan Blezek)<br />
#[[2008_Winter_Project_Week:CPack |CPack]], CTest, CMake infrastructure Improvements(Katie Hayes, Steve Pieper, Bill Hoffman, Sebastien Barre, Will Schroeder)<br />
#[[2008_Winter_ProjectWeek:SlicerCookbook | Slicer GUI Style Guide and Cookbook]] (Wendy Plesniak, Sebastien Barre)<br />
#[[2008_Winter_Project_Week_VolumeRendering|Volume rendering]] (Andy Freudling, Steve Pieper, Benjamin Grauer) <br />
#[[2008_Winter_Project_Week_VolumeRenderingUsingCuda|Volume rendering using Cuda]] (Benjamin Grauer, Nobuhiko Hata) <br />
#[[2008_Winter_Project_Week:3DWWidgets |3D W Widgets and Picking]] (Will Schroeder, Nicole Aucoin, Curt Lisle, Kiran Shivana)<br />
#[[2008_Winter_Project_Week:XNAT_Integration | XNAT Integration]] (Dan Marcus, Steve Pieper, Stephen Aylward, Jeff Grethe, Julien Jomier, Kevin Archie, Misha Milchenko) <br />
#[[2008_Winter_Project_Week:KWWidgets | KWWidgets Roadmap]] (Sebastien Barre, Wendy Plesniak, Katie Hayes)<br />
<br />
===External Collaborations===<br />
#[[2008_Winter_Project_Week:Astronomical_Medicine|Astronomical coordinate system support]] [Harvard IIC] (Mike Halle, Douglas Alan, Steve Pieper)<br />
#[[2008_Winter_Project_Week:Meshing Techniques into NA-MIC Toolkit|New Meshing Techniques into NA-MIC]] [Univ. of Iowa] (Vince Magnotta and Nicole Grosland, )<br />
#[[2008_Winter_Project_Week:Finite Element Meshing into NA-MIC|Meshing Workflow into Slicer]] [Univ. of Iowa] (Curt Lisle, Nicole Grosland, Vince Magnotta, Kiran Shivana, Steve Pieper, Brad Davis)<br />
#[[2008_Winter_Project_Week:SmallAnimalEvalNCI |Evaluating NA-MIC Tools for Small Animal Imaging Workflows]] [NCI] (Curt Lisle, Jack Collins, Killian Pohl)<br />
#[[2008_Winter_Project_Week:IGT_IGSTK_Slicer| IGSTK-Slicer]] [Georgetown] (Haiying Liu, Patrick Cheng, Noby Hata, Junichi Tokuda, Luis Ibanez, Steve Pieper)<br />
#[[2008_Winter_Project_Week:IGT_Intelligent_Surgical_Instrument_Projects| Japanese Intelligent Surgical Instrument Project]] [AIST] (Noby Hata, Chinzei, Hong)<br />
#[[2008_Winter_Project_Week:GoFigure |GoFigure:High-Level Microscopy Image analysis Application and Algorithms]] [CalTech-Harvard Medical School] (Alex G, Sean Megason, Arnaud Gelas?, Stephen Aylward)<br />
#[[2008_Winter_Project_Week:microslicer_3 | Doing microscopy image analysis with Slicer3]] [The Ohio State University](Kishore Mosaliganti, Raghu Machiraju, Brad Davis,Stephen Aylward, Steve Pieper)<br />
#[[2008_Winter_Project_Week:fmri_image_analysis | fMRI Analysis with Slicer 3]] [The Ohio State University](Firdaus Janoos,Raghu Machiraju, Luis Ibanez,Steve Pieper, Wendy Plesniak)<br />
#[[2008_Winter_Project_Week:Resampling_DTIs_with_Slicer3|Resampling DTIs with Slicer 3]] (Francois Budin, Sylvain Bouix, Xiaodong Tao)<br />
#[[2008_Winter_Project_Week:fmri_eeg_analysis | Analyzing fMRI and concurrent EEG with Slicer and SCIRun]] [BWH] (Padma Sundaram)<br />
#[[2008_Winter_Project_Week:fmri_var | Bayesian hierarchical models for fMRI variance components analysis]] (Kinh Tieu, Sandy Wells, Jim Miller)<br />
<br />
== Dates.Venue.Registration ==<br />
<br />
'''Dates:''' <br />
* The All Hands Meeting and External Advisory Board Meeting will be held on '''Thursday, January 10th'''. <br />
* Project Activities will be held rest of the week between '''Monday, January 7th and Friday, January 11th'''.<br />
<br />
'''Venue:''' The venue for the meeting is [http://www.marriott.com/hotels/travel/slccc-salt-lake-city-marriott-city-center/ Marriot City Center, Salt Lake City, Utah] Mariott City Center, Salt Lake City, Utah. [http://marriott.com/property/meetingsandevents/floorplans/slccc (Floorplan)]. To reserve rooms at the meeting rate of $129/night, please call the hotel at 1-801-961-8700 or 1-866-961-8700 (toll free) and mention that you are attending the NAMIC meeting. Please note that we do need attendees to use this hotel in order to not incur additional charges for the use of conference rooms.<br />
<br />
<big> '''Registration:''' We are charging a registration fee to all participants. The fee covers the costs of the facilities and food provided. In order to keep the fee low, we need to get a sufficient number of hotel nights by our participants. See above for more on this. Please click [http://www.sci.utah.edu/namic2008/registration.html '''here'''] for online registration. This registration must be completed by Friday, December 14, 2007. </big><br />
<br />
<br />
Please note that this information can also be found [[AHM_2008#Dates.Venue.Registration|here.]]<br />
<br />
([[Project Week Logistics Checklist|This is a checklist for the onsite planning items]])<br />
<br />
=== Agenda===<br />
<br />
[[AHM_2008#Agenda|Agenda for AHM 2008 and Project Week]]<br />
<br />
=== Preparation ===<br />
<br />
# Please make sure that you are on the [http://public.kitware.com/cgi-bin/mailman/listinfo/na-mic-project-week na-mic-project-week mailing list]<br />
# [[Engineering:TCON_2007#2007-11-29|November 29, 2007: Kickoff TCON#1 (w/ NA-MIC Engeering Core only) to discuss Projects and Assign/Verify Teams]]<br />
# [[Engineering:TCON_2007#2007-12-06|December 6, 2007: TCON#2 with all participants to Assign/Verify Teams]]<br />
# [[Engineering:TCON_2007#2007-12-13|December 13, 2007: TCON#3 with Breakout Session owners to review agendas]]<br />
# [[Engineering:TCON_2007#2007-12-13|December 20, 2007: TCON#4 to discuss outstanding projects and teams]]<br />
# By December 20, 2008: [[2008_Winter_Project_Week_Template|Complete a templated wiki page for your project]]. Please do not edit the template page itself, but create a new page for your project and cut-and-paste the text from this template page. If you have questions, please send an email to tkapur at bwh.harvard.edu.<br />
# [[Engineering:TCON_2007#2008-01-03|January 3, 2008: TCON#5 to discuss outstanding projects and teams]]<br />
# January 3, 2008: Create a directory for each project on the [[Engineering:SandBox|NAMIC Sandbox]] (Zack)<br />
##[https://www.kitware.com/Admin/SendPassword.cgi Ask Zack for a Sandbox account]<br />
## Commit on each sandbox directory the code examples/snippets that represent our first guesses of appropriate methods. (Luis and Steve will help with this, as needed)<br />
## Gather test images in any of the Data sharing resources we have (e.g. the BIRN). These ones don't have to be many. At least three different cases, so we can get an idea of the modality-specific characteristics of these images. Put the IDs of these data sets on the wiki page. (the participants must do this.)<br />
## Setup nightly tests on a separate Dashboard, where we will run the methods that we are experimenting with. The test should post result images and computation time. (Zack)<br />
# Please note that by the time we get to the project event, we should be trying to close off a project milestone rather than starting to work on one...<br />
<br />
== Previous Project Events ==<br />
<br />
A history of all the programming/project events in NA-MIC is available by following [[Project Events|this link]].</div>Serdarhttps://www.na-mic.org/w/index.php?title=2008_Winter_Project_Week_Image_Registration_Update&diff=202992008 Winter Project Week Image Registration Update2007-12-28T19:44:25Z<p>Serdar: /* Attendees */</p>
<hr />
<div>__NOTOC__<br />
Back to [[AHM_2008]]<br />
<br />
<br />
=Breakout session on "Registration methods in the NAMIC Toolkit"=<br />
<br />
== Attendees ==<br />
<br />
<em>Please add your name below, if your plan on attending.</em><br />
<br />
* Guido Gerig<br />
* Jim Miller<br />
* Stephen Aylward<br />
* Ross Whitaker<br />
* Luis Ibanez<br />
* Alex hanfei Gouaillard<br />
* Casey Goodlett<br />
* Bill Lorensen<br />
* Luca Antiga<br />
* Clement Vachet<br />
* Ron Kikinis<br />
* Serdar Balci<br />
<br />
== Topics ==<br />
<br />
<em>Please submit your ideas for topics to be discussed.</em><br />
<br />
* Demonstration of a registration module in Slicer (Stephen)<br />
* Needs (Guido)<br />
* What is available (Guido)<br />
* Use in a pipeline (Guido)<br />
* Validation (Guido)<br />
* Testing (Luis)<br />
* Heuristics for selecting components(Luis)<br />
<br />
== Presentations ==<br />
<br />
<em>If you would like to give a short (5-10 minute) presentation, please describe it below.</em><br />
<br />
* Opening remarks (Guido)<br />
* Demonstration of a registration module in Slicer (Stephen)<br />
* Discussion of directions for future work (assign priorities and responsibilities)<br />
** Passing and tracking transforms in Slicer (Jim)<br />
** Regularization of BSplines (Stephen)<br />
** Adding "don't process" zones in BSplines (Stephen)<br />
** Objects as masks: mrml -=> SpatialObject conversion (Stephen)<br />
** Reducing BSpline memory requirements (Brad)<br />
** Specializing particular metric/transform combinations (Luis)<br />
** Batch processing registration<br />
*** Atlas formation<br />
*** Atlas application<br />
** Validation<br />
*** STAPLE<br />
<br />
== Related Links ==<br />
<br />
<em>Add papers, presentations, wiki pages, etc.</em><br />
<br />
* [http://www.vtk.org/Wiki/Proposals:GridComputing Grid Computing in ITK]<br />
* [http://www.na-mic.org/Wiki/index.php/ITK_Registration_Optimization ITK Multi-threaded registration]<br />
* [http://www.vtk.org/Wiki/ITK_Roadmap_2007_2008 ITK Roadmap for 2008]</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=19592Projects:GroupwiseRegistration2007-12-14T22:24:40Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
''Objective Function''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
''Deformation Model''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
''Implementation''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available [http://www.na-mic.org/svn/NAMICSandBox/trunk/MultiImageRegistration/ (code)].<br />
<br />
''Results''<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average over the whole population.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17738Projects:GroupwiseRegistration2007-11-16T17:22:11Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit(ITK)<br />
and made the implementation publicly available.<br />
<br />
'''Results'''<br />
<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17628Projects:GroupwiseRegistration2007-11-13T15:05:47Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around 32 voxels.<br />
We increase the resolution of the deformation field to 8 voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available. <br />
We run experiments using a dataset of 50 MR images <br />
with 256x256x128 voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Results'''<br />
<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
For each image in the dataset, an automatic tissue classification<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17626Projects:GroupwiseRegistration2007-11-13T15:00:56Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around 32 voxels.<br />
We increase the resolution of the deformation field to 8 voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available. <br />
We run experiments using a dataset of 50 MR images <br />
with 256x256x128 voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|350px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|350px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with 256x256x128 voxels <br />
and 0.9375x0.9375x1.5 mm<sup>3</sup> spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17566Projects:GroupwiseRegistration2007-11-09T23:21:58Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is <math>l</math>'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around 32 voxels.<br />
We increase the resolution of the deformation field to 8 voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17565Projects:GroupwiseRegistration2007-11-09T23:20:15Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is $l$'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|350px|Figure 4: An example showing the multi-resolution scheme. The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. Also note that the objective function is only evaluated on a small subset of input points. ]]<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17563Projects:GroupwiseRegistration2007-11-09T23:17:48Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is $l$'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
'''Implementation'''<br />
<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|350px|Figure 3: A registration schedule using gradually increasing deformation field complexity. From left to right deformation fields for increasing deformation field complexity is shown. ]]<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17558Projects:GroupwiseRegistration2007-11-09T23:14:58Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
[[Image:GroupwiseBspline.png|thumb|350px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
<br />
<br />
where <math>B_l</math> is $l$'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17556Projects:GroupwiseRegistration2007-11-09T23:12:45Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
[[Image:GroupwiseBspline.png|thumb|200px|Figure 2: An example deformation field. The local neighborhood affecting the deformation is overlayed on the image.]]<br />
<br />
where <math>B_l</math> is $l$'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17554Projects:GroupwiseRegistration2007-11-09T23:10:02Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
:<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
:<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
[[Image:GroupwiseBspline.png|thumb|200px|Figure 1: desc]]<br />
<br />
where $B_l$ is $l$'th cubic B-spline basis function. <math>(u,v,w)</math> is the distance <br />
to <math>(x,y,z)</math> from the control point <math>\Phi_{i,j,k}</math> as shown in Figure 2.<br />
<br />
The deformation of a given point can be found using only the control points in the neighborhood of the given point. Therefore,<br />
optimization of the objective function can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17552Projects:GroupwiseRegistration2007-11-09T23:03:19Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<br />
<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<br />
<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
[[Image:GroupwiseBspline.png|thumb|200px|Figure 1: desc]]<br />
<br />
where <br />
$\mathbf{x}=(x,y,z)$,<br />
$i=\lfloor x/n_x \rfloor -1$, <br />
$j=\lfloor y/n_y\rfloor -1$, <br />
$k=\lfloor z/n_z\rfloor -1$, <br />
$u=x/n_x -\lfloor x/n_x\rfloor$, <br />
$v=y/n_y -\lfloor y/n_y\rfloor$, <br />
$w=z/n_z -\lfloor z/n_z\rfloor$<br />
and where $B_l$ is $l$'th cubic B-spline basis function. <br />
Using the same expressions for $u,v$ and $w$ as above, the derivative of the deformation field with respect to B-spline coefficients <br />
can be given by<br />
<math><br />
\frac{\partial T_{local}(x,y,z) }{\partial \Phi_{i,j,k}} = B_l(u)B_m(v)B_n(w)<br />
</math><br />
where $l=i-\lfloor x/n_x\rfloor+1$, $m=j-\lfloor y/n_y\rfloor+1$ and $n=k-\lfloor z/n_z\rfloor+1$. We consider $B_l(u) = 0$ for $l<0 $ and $l>3$.<br />
The derivative terms are nonzero only in the neighborhood of a given point. Therefore,<br />
optimization of the objective function using gradient descent can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17551Projects:GroupwiseRegistration2007-11-09T23:02:14Z<p>Serdar: /* Description */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical region. On the left is shown the Gaussian(blue) fittet to <br />
a real sample from the dataset we used along with the non-parametric density estimate(red). <br />
Note that the distribution is bi-modal because of white matter-gray matter transaction.]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
Following Rueckert et al.'s formulation \cite{rueckert},<br />
we let $\mathbf{\Phi}$ denote an<br />
$n_x \times n_y \times n_z $ grid of control points $\Phi_{i,j,k}$ with uniform<br />
spacing. The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
[[Image:GroupwiseBspline.png|thumb|200px|Figure 1: desc]]<br />
<br />
where <br />
$\mathbf{x}=(x,y,z)$,<br />
$i=\lfloor x/n_x \rfloor -1$, <br />
$j=\lfloor y/n_y\rfloor -1$, <br />
$k=\lfloor z/n_z\rfloor -1$, <br />
$u=x/n_x -\lfloor x/n_x\rfloor$, <br />
$v=y/n_y -\lfloor y/n_y\rfloor$, <br />
$w=z/n_z -\lfloor z/n_z\rfloor$<br />
and where $B_l$ is $l$'th cubic B-spline basis function. <br />
Using the same expressions for $u,v$ and $w$ as above, the derivative of the deformation field with respect to B-spline coefficients <br />
can be given by<br />
<math><br />
\frac{\partial T_{local}(x,y,z) }{\partial \Phi_{i,j,k}} = B_l(u)B_m(v)B_n(w)<br />
</math><br />
where $l=i-\lfloor x/n_x\rfloor+1$, $m=j-\lfloor y/n_y\rfloor+1$ and $n=k-\lfloor z/n_z\rfloor+1$. We consider $B_l(u) = 0$ for $l<0 $ and $l>3$.<br />
The derivative terms are nonzero only in the neighborhood of a given point. Therefore,<br />
optimization of the objective function using gradient descent can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17550Algorithm:MIT:New2007-11-09T22:59:13Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We extend a previously demonstrated entropy based groupwise registration method to include a free-form deformation model based on B-splines. We provide an efficient implementation using stochastic gradient descents in a multi-resolution setting. We demonstrate the method in application to a set of 50 MRI brain scans and compare the results to a pairwise approach using segmentation labels to evaluate the quality of alignment.<br />
[[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
<!--<br />
|-<br />
<br />
| | [[Image:brain.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<!--<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
--><br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<!--<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|center|200px]]<br />
| | <br />
<br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<!--<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.<br />
--></div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17548Algorithm:MIT:New2007-11-09T22:58:07Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We extend a previously demonstrated entropy based groupwise registration method to include a free-form deformation model based on B-splines. We provide an efficient implementation using stochastic gradient descents in a multi-resolution setting. We demonstrate the method in application to a set of 50 MRI brain scans and compare the results to a pairwise approach...<br />
[[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
<!--<br />
|-<br />
<br />
| | [[Image:brain.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<!--<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|center|200px]]<br />
| | <br />
<br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<!--<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.<br />
--></div>Serdarhttps://www.na-mic.org/w/index.php?title=Projects:GroupwiseRegistration&diff=17546Projects:GroupwiseRegistration2007-11-09T22:57:33Z<p>Serdar: /* Key Investigators */</p>
<hr />
<div>Back to [[NA-MIC_Collaborations|NA-MIC_Collaborations]], [[Algorithm:MIT|MIT Algorithms]]<br />
<br />
= Non-rigid Groupwise Registration =<br />
<br />
We aim at providing efficient groupwise registration algorithms <br />
for population analysis of anatomical structures.<br />
Here we extend a previously demonstrated entropy based groupwise registration method <br />
to include a free-form deformation model based on B-splines. <br />
We provide <br />
an efficient implementation using stochastic gradient descents <br />
in a multi-resolution setting. <br />
We demonstrate the method in application to a set of 50 MRI brain scans <br />
and compare the results to a pairwise approach <br />
using segmentation labels to evaluate the quality of alignment.<br />
Our results indicate that increasing the complexity of the deformation model<br />
improves registration accuracy significantly, especially at cortical regions.<br />
<br />
= Description =<br />
<br />
We first describe <br />
the stack entropy cost function and the B-spline based deformation model.<br />
Then we discuss implementation details. <br />
Next, we compare groupwise registration to the pairwise method and<br />
evaluate both methods using label prediction values.<br />
<br />
<br />
<br />
<br />
<br />
'''Objective Function'''<br />
<br />
[[Image:GroupwiseStackBiModal.PNG|thumb|350px|Figure 1: On the left is shown a stack of images <br />
and a sample pixel stack around a cortical ]]<br />
<br />
In order to align all subjects in the population,<br />
we consider sum of pixelwise entropies as a joint alignment criterion.<br />
The justification for this approach is that if the images are aligned properly, <br />
intensity values at corresponding coordinate locations from all the images <br />
will form a low entropy distribution.<br />
This approach does not require the use of a reference subject; all<br />
subjects are simultenously driven to the common tendency of the population.<br />
<br />
We employ a kernel based density estimation scheme to estimate univariate entropies.<br />
Using the entropy measure we obtain a better treatment of transitions between different<br />
tissue types, such as gray matter-white matter transitions in the cortical regions<br />
where intensity distributions can be bi-modal as shown in Figure 1.<br />
<br />
<br />
<br />
'''Deformation Model'''<br />
<br />
For the nonrigid deformation model,<br />
we define a combined transformation consisting of <br />
a global and a local component<br />
<math><br />
T(\mathbf{x}) = T_{local}({T_{global}(\mathbf{x})})<br />
</math><br />
<br />
where <math>T_{global}</math> is a twelve parameter affine transform and <br />
<math>T_{local}</math> is a deformation model based on B-splines.<br />
<br />
Following Rueckert et al.'s formulation \cite{rueckert},<br />
we let $\mathbf{\Phi}$ denote an<br />
$n_x \times n_y \times n_z $ grid of control points $\Phi_{i,j,k}$ with uniform<br />
spacing. The free form deformation can be written as the 3-D tensor product<br />
of 1-D cubic B-splines.<br />
<math><br />
T_{local}(\mathbf{x}) = \mathbf{x} + \sum_{l=0}^3\sum_{m=0}^3\sum_{n=0}^3 B_l(u)B_m(v)B_n(w) \Phi_{i+l,j+m,k+n}<br />
</math><br />
<br />
[[Image:GroupwiseBspline.png|thumb|200px|Figure 1: desc]]<br />
<br />
where <br />
$\mathbf{x}=(x,y,z)$,<br />
$i=\lfloor x/n_x \rfloor -1$, <br />
$j=\lfloor y/n_y\rfloor -1$, <br />
$k=\lfloor z/n_z\rfloor -1$, <br />
$u=x/n_x -\lfloor x/n_x\rfloor$, <br />
$v=y/n_y -\lfloor y/n_y\rfloor$, <br />
$w=z/n_z -\lfloor z/n_z\rfloor$<br />
and where $B_l$ is $l$'th cubic B-spline basis function. <br />
Using the same expressions for $u,v$ and $w$ as above, the derivative of the deformation field with respect to B-spline coefficients <br />
can be given by<br />
<math><br />
\frac{\partial T_{local}(x,y,z) }{\partial \Phi_{i,j,k}} = B_l(u)B_m(v)B_n(w)<br />
</math><br />
where $l=i-\lfloor x/n_x\rfloor+1$, $m=j-\lfloor y/n_y\rfloor+1$ and $n=k-\lfloor z/n_z\rfloor+1$. We consider $B_l(u) = 0$ for $l<0 $ and $l>3$.<br />
The derivative terms are nonzero only in the neighborhood of a given point. Therefore,<br />
optimization of the objective function using gradient descent can be implemented efficiently.<br />
<br />
<br />
[[Image:GroupwiseMultiResolution.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseIncreasingScale.PNG|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseMeanImages.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsWMGM.png|thumb|200px|Figure 1: desc]]<br />
[[Image:GroupwiseBarsManual.png|thumb|200px|Figure 1: desc]]<br />
<br />
<br />
%The resolution of the deformation field<br />
%can be controlled by the grid size.<br />
%In order to capture shape variations at different resolution levels <br />
%we start the registration at a low resolution level with a coarse grid<br />
%and increase the resolution by refining the grid of control points.<br />
%Results of a registration at a coarse level are used to initialize <br />
%the grid of control points at a higher resolution level.<br />
<br />
As none of the images are chosen as an anatomical reference,<br />
it is necessary to add a geometric constraint to define the reference coordinate frame.<br />
Similar to Bhatia et al. \cite{bhatia}, we define the reference frame<br />
by constraining the average deformation to be the identity transform:<br />
<math><br />
\frac{1}{N}\sum_{n=1}^{N} T_n(\mathbf{x}) = \mathbf{x}<br />
</math><br />
<br />
This constraint assures that the reference frame lies in the center of the <br />
population. In the case of B-splines, the constraint can be satisfied <br />
by constraining the sum of B-spline coefficients across images<br />
to be zero. In the gradient descent optimization scheme,<br />
the constraint can be forced by<br />
subtracting the mean from each update vector \cite{bhatia}.<br />
<br />
<br />
<br />
<br />
<br />
'''Implementation'''<br />
<br />
We provide an efficient optimization scheme by using line search with the gradient descent algorithm.<br />
For computational efficiency, we employ a stochastic subsampling procedure \cite{pluim}. <br />
In each iteration of the algorithm, <br />
a random subset is drawn from all samples and the objective function is evaluated<br />
only on this sample set. <br />
The number of samples to be used in this method depends on the number of the parameters of<br />
the deformation field. <br />
Using the number of samples on the order of the number of variables works well in practice.<br />
<br />
To obtain a dense deformation field capturing anatomical variations at different scales,<br />
we gradually increase the complexity of the deformation field by refining the grid of B-spline control points.<br />
First, we perform a global registration with affine transforms. <br />
Then, we use affine transforms to initialize a low resolution deformation field <br />
at a grid spacing around $32$~voxels.<br />
We increase the resolution of the deformation field to $8$~voxels<br />
by using registration results at coarser grids to initialize finer grids.<br />
<br />
As in every iterative search algorithm, local minima pose a significant problem. <br />
To avoid local minima we use a multi-resolution optimization scheme for each resolution level of the deformation field.<br />
The registration is first performed at a coarse scale by downsampling the input.<br />
Results from coarser scales are used to initialize<br />
optimization at finer scales. <br />
For each resolution level of the deformation field we used a multi-resolution scheme of three <br />
image resolution levels.<br />
<br />
We implemented our groupwise registration method in a multi-threaded fashion using Insight Toolkit (ITK)<br />
and made the implementation publicly available \cite{itk}. <br />
We run experiments using a dataset of 50 MR images <br />
with $256\times256\times128$ voxels on a workstation with four CPUs and <br />
8GB of memory. The running time of the algorithm <br />
is about 20 minutes for affine registration and two days for non-rigid registration of the entire dataset.<br />
The memory requirement of the algorithm depends linearly on the number of input images <br />
and was around 3GB for our dataset.<br />
<br />
<br />
<br />
'''Progress'''<br />
<br />
We tested the groupwise registration algorithm on a MR brain dataset and<br />
compared the results to a pairwise approach \cite{joshi}. <br />
The dataset consists of 50 MR brain images of three subgroups:<br />
schizophrenics, affected disorder and normal control patients. <br />
MR images are T1 scans with $256\times256\times128$ voxels <br />
and $0.9375\times0.9375\times1.5$~$\mbox{mm}^3$ spacing. <br />
MR images are preprocessed by skull stripping.<br />
To account for global intensity variations, images are normalized<br />
to have zero mean and unit variance.<br />
For each image in the dataset, an automatic tissue classification \cite{pohl}<br />
was performed, yielding gray matter (GM), white matter (WM) and cerebro-spinal<br />
fluid (CSF) labels. In addition, manual segmentations of four subcortical regions <br />
(left and right hippocampus and amygdala) and four cortical regions (left and right superior temporal <br />
gyrus and para-hippocampus) were available for each MR image.<br />
<br />
<br />
The prediction accuracy reported in Table \ref{table:pred} is lower than what is typically<br />
achieved by segmentation methods. This is to be expected as our<br />
relatively simple label prediction method only considers voxelwise<br />
majority of the labels in the population and does not use the novel<br />
image intensity to predict the labels. Effectively, Table \ref{table:pred} reports<br />
the accuracy of the spatial prior (atlas) in predicting the labels<br />
before the local intensity is used to refine the segmentation.<br />
Increasing the complexity of the deformation model improves the<br />
accuracy of prediction. An interesting open problem is automatically<br />
identifying the appropriate deformation complexity before the<br />
registration overfits and the accuracy of prediction goes down. We<br />
also note that the alignment of the subcortical structures is much<br />
better than that of the cortical regions. It is not surprising as the<br />
registration algorithm does not use the information about geometry of the cortex<br />
to optimize the alignment of the cortex. In addition, it has<br />
been often observed that the cortical structures exhibit higher<br />
variability across subjects when considered in the 3D volume rather<br />
than modelled on the surface.<br />
<br />
Our experiments highlight the need for further research in developing<br />
evaluation criteria for image alignment. We used the standard Dice<br />
measure, but it is not clear that this measurement captures all the<br />
nuances of the resulting alignment.<br />
<br />
Comparing the groupwise registration to the pairwise approach, we<br />
observe that the sharpness of the mean images and the tissue overlaps<br />
in Figure~\ref{fig:mean} look visually similar. From Table~\ref{table:pred}, we note that<br />
groupwise registration performs slightly better than the pairwise<br />
setting in most of the cases, especially as we increase the complexity<br />
of the warp. This suggests that considering the population as a whole<br />
and registering subjects jointly brings the population into better<br />
alignment than matching each subject to a mean template<br />
image. However, the advantage shown here is only slight; more<br />
comparative studies are needed of the two approaches.<br />
<br />
%Average Dice measures in Table \ref{table:pred} indicate relatively low values, especially for<br />
%manual segmentations of cortical structures. <br />
%This is to be expected as our relatively simple label prediction method<br />
%only considers voxelwise majority of the labels in the population. <br />
%Higher overlap values can be obtained by using more advanced segmentation techniques which make us of the <br />
% shapes of the structures.<br />
%As our method keeps the voxelwise intensity distributions in the population, these distributions<br />
%can be used to supply prior information to a more advanced segmentation algorithm to increase <br />
%the segmentation accuracy.<br />
<br />
<br />
%Evaluation of a groupwise registration<br />
%challenging task as it is hard to distinguish between registration inaccuracy and anatomical variability <br />
%by considering Dice measure as the only evaluation criterion.<br />
%However, comparing overlap measures relative to each other gives us useful information.<br />
%The blurriness of the mean images in figure \ref{fig:mean} indicate high anatomical variabilities<br />
%at cortical structures. <br />
%Part of this anatomical variability is captured by higher resolution levels of B-splines, <br />
%as can be observed from the increase in Dice measures for cortical structures in Table \ref{table:pred}.<br />
%White matter and grey matter tissue type have a significant component in cortical regions; therefore,<br />
%the overlap measures for these tissue types also increase as the resolution of the deformation field increases.<br />
%Label overlap values for subcortical structures do not improve significantly with an increase in the resolution of the <br />
%deformation field.<br />
%These structures do not highly correlate with local intensity variations; therefore,<br />
%we believe that <br />
%the low Dice measures are mostly due to anatomical variabilities in these structures.<br />
<br />
<br />
<br />
%As for the comparison of groupwise to pairwise approach, <br />
%we can observe that the sharpness of the mean images and the tissue overlaps in figure \ref{fig:mean} look visually similar. <br />
%From figure \ref{table:pred} we note that groupwise registration performs <br />
%slightly better than the pairwise setting in most of the cases. <br />
%This suggests that considering the population as a whole and registering subjects jointly brings the population into better %alignment than matching each subject to a mean template image.<br />
%A mean template image cannot fully explain multi-modal distributions; our groupwise setting<br />
%captures the modes in the distributions by considering stack entropies as an alignment criterion.<br />
<br />
<br />
<br />
We compare our groupwise algorithm to a pairwise method where we register<br />
each subject to the mean intensity using sum of square differences.<br />
The objective function for pairwise registration to the mean can be described as follows<br />
<math><br />
f_{pair} = \sum_{n=1}^{N} ( I_n(T_n(x)) - \mu(x) )^2<br />
</math><br />
where $\mu$ is defined as the mean of the intensities $\mu(x) = \frac{1}{N}\sum_{n=1}^{N} I_n(T_n(x))$. <br />
During each iteration we consider the mean image as a reference image<br />
and register<br />
every subject to the mean image using sum of squared differences. <br />
After each iteration the mean image is updated and pairwise registrations are performed until convergence.<br />
<br />
<br />
<br />
The images in Figure \ref{fig:mean} show central slices of 3D images after registration. <br />
Visually, mean images get sharper and variance images becomes darker, especially around central ventricles and cortical regions. <br />
We can observe that anatomical variability at cortical regions causes significant blur for <br />
GM, WM and CSF structures using affine registration. <br />
Finer scales of B-spline deformation fields capture a significant part of this anatomical variability and <br />
the tissue label overlap images get sharper.<br />
<br />
<br />
<br />
We evaluate registration results by measuring label prediction accuracy<br />
in a leave-one-out fashion for the two different sets of segmentation labels.<br />
To predict the segmentation labels of a subject, we <br />
use majority voting for the labels in the rest of the population. <br />
We compute the Dice measure between the predicted and the true labels and average<br />
over the whole population.<br />
%Table \ref{table:pred} shows average Dice measures for two automatically segmented <br />
%tissue types: grey matter (GM) and white matter (WM) in the first two columns.<br />
%We excluded cerebro-spinal fluid (CSF) from the evaluation because CSF labels <br />
%near the cortex get corrupted<br />
%after preprocessing the data with skull stripping.<br />
%In the last two columns of Table \ref{table:pred} we display average prediction values for <br />
%manual segmentations of the cortical structures: superior temporal gyrus and para-hippocampus in the <br />
%left and right hemispheres; the subcortical structures: amygdala and hippocampus in the left and <br />
%right hemispheres.<br />
<br />
= Key Investigators =<br />
<br />
* S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton.<br />
<br />
= Publications =<br />
<br />
* S.K. Balci, P. Golland, M.E. Shenton, W.M. Wells III. Free-Form B-spline Deformation Model for Groupwise Registration. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 23-30, 2007.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17544Algorithm:MIT:New2007-11-09T22:55:52Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We extend a previously demonstrated entropy based groupwise registration method to include a free-form deformation model based on B-splines. We provide an efficient implementation using stochastic gradient descents in a multi-resolution setting. We demonstrate the method in application to a set of 50 MRI brain scans and compare the results to a pairwise approach...<br />
[[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
<!--<br />
|-<br />
<br />
| | [[Image:brain.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17543Algorithm:MIT:New2007-11-09T22:54:44Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We extend a previously demonstrated entropy based groupwise registration method to include a free-form deformation model based on B-splines. We provide an efficient implementation using stochastic gradient descents in a multi-resolution setting. We demonstrate ... method[[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
<!--<br />
|-<br />
<br />
| | [[Image:brain.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
--><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17541Algorithm:MIT:New2007-11-09T22:52:31Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We are exploring algorithms for groupwise registration of medical data. [[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> S.K. Balci, P. Golland, S. Wells, M.R. Sabuncu, S. Bouix, M. Shenton. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|center|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17539Algorithm:MIT:New2007-11-09T22:50:35Z<p>Serdar: /* Groupwise Registration */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We are exploring algorithms for groupwise registration of medical data. [[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> Serdar K Balci, Polina Golland, Sandy Wells, Mert R Sabuncu, Martha Shenton, . Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:mit_fmri_clustering_parcellation2_xsub.png|center|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17536Algorithm:MIT:New2007-11-09T22:47:08Z<p>Serdar: /* MIT Projects */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We are exploring algorithms for groupwise registration of medical data. [[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> Serdar K Balci, Polina Golland, Sandy Wells, Lilla Zollei, Mert R Sabuncu and Kinh Tieu. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:Brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17535Algorithm:MIT:New2007-11-09T22:45:47Z<p>Serdar: /* MIT Projects */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We are exploring algorithms for groupwise registration of medical data. [[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> Serdar K Balci, Polina Golland, Sandy Wells, Lilla Zollei, Mert R Sabuncu and Kinh Tieu. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|Center|200px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:Brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdarhttps://www.na-mic.org/w/index.php?title=Algorithm:MIT:New&diff=17534Algorithm:MIT:New2007-11-09T22:44:19Z<p>Serdar: /* MIT Projects */</p>
<hr />
<div>Back to [[Algorithm:Main|NA-MIC Algorithms]]<br />
<br />
= Overview of MIT Algorithms =<br />
<br />
A brief overview of the MIT's algorithms goes here. This should not be much longer than a paragraph. Remember that people visiting your site want to be able to understand very quickly what you're all about and then they want to jump into your site's projects. The projects below are organized into a two column table: the left column is for representative images and the right column is for project overviews. The number of rows corresponds to the number of projects at your site. Put the most interesting and relevant projects at the top of the table. You do not need to organize the table according to subject matter (i.e. do not group all segmentation projects together and all DWI projects together).<br />
<br />
[[#Shape_Based_Segmentation_and_Registration|Shape Based Segmentation and Registration]]<br />
<br />
[[#Effects_of_Registration_Regularization_on_Segmentation_Accuracy|Effects of Registration Regularization on Segmentation Accuracy]]<br />
<br />
[[#Multimodal_Atlas|Multimodal Atlas]]<br />
<br />
[[#Shape_Analysis_With_Overcomplete_Wavelets|Shape Analysis With Overcomplete Wavelets]]<br />
<br />
[[#fMRI_clustering|fMRI clustering]]<br />
<br />
[[#Joint_Registration_and_Segmentation_of_DWI_Fiber_Tractography|Joint Registration and Segmentation of DWI Fiber Tractography]]<br />
<br />
[[#Shape_Based_Level_Segmentation|Shape Based Level Segmentation]]<br />
<br />
[[#DTI_Fiber_Clustering_and_Fiber-Based_Analysis|DTI Fiber Clustering and Fiber-Based Analysis]]<br />
<br />
[[#Fiber_Tract_Modeling.2C_Clustering.2C_and_Quantitative_Analysis|Fiber Tract Modeling, Clustering and Quantitative Analysis]]<br />
<br />
[[#DTI-based_Segmentation|DTI-based Segmentation]]<br />
<br />
[[#Stochastic_Tractography|Stochastic Tractography]]<br />
<br />
[[#fMRI_Detection_and_Analysis|fMRI Detection and Analysis]]<br />
<br />
[[#Population_Analysis_of_Anatomical_Variability|Population Analysis of Anatomical Variability]]<br />
<br />
[[#Groupwise_Registration|Groupwise Registration]]<br />
<br />
= MIT Projects =<br />
<br />
<br />
{|<br />
| style="width:10%" | [[Image:Progress_Registration_Segmentation_Shape.jpg|left|200px]]<br />
| style="width:90%" | <br />
<br />
== [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|Shape Based Segmentation and Registration]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:JointRegSeg.png|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:RegistrationRegularization|Optimal Atlas Regularization in Image Segmentation]] ==<br />
<br />
We propose a unified framework for computing atlases from manually labeled data sets at various degrees of “sharpness” and the joint registration and segmentation of a new brain with these atlases. Using this framework, we investigate the tradeoff between warp regularization and image fidelity, i.e. the smoothness of the new subject warp and the sharpness of the atlas in a segmentation application.<br />
[[Algorithm:MIT:RegistrationRegularization|More...]]<br />
<br />
<font color="red">'''New:'''</font> B.T.T. Yeo, M.R. Sabuncu, R. Desikan, B. Fischl, P. Golland. Effects of Registration Regularization and Atlas Sharpness on Segmentation Accuracy. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 683-691, 2007. '''MICCAI Young Scientist Award.'''<br />
<br />
|-<br />
<br />
| | [[Image:ICluster_templates.gif|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Multimodal Atlas |Multimodal Atlas]] ==<br />
<br />
In this work, we propose and investigate an algorithm that jointly co-registers a collection of images while computing multiple templates. The algorithm, called '''iCluster''', is used to compute multiple atlases for a given population.<br />
[[Algorithm:MIT:Multimodal Atlas|More...]]<br />
<br />
<font color="red">'''New: '''</font> M.R. Sabuncu, M.E. Shenton, P. Golland. Joint Registration and Clustering of Images. In Proceedings of MICCAI 2007 Statistical Registration Workshop: Pair-wise and Group-wise Alignment and Atlas Formation, 47-54, 2007.<br />
[[Algorithm:MIT:Multimodal Atlas#Publications|More...]]<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:GroupwiseSummary.PNG|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Groupwise_Registration#Introduction|Groupwise Registration]] ==<br />
<br />
We are exploring algorithms for groupwise registration of medical data. [[Algorithm:MIT:Groupwise_Registration|More...]]<br />
<br />
<font color="red">'''New:'''</font> Serdar K Balci, Polina Golland, Sandy Wells, Lilla Zollei, Mert R Sabuncu and Kinh Tieu. Groupwise registration of medical data.<br />
<br />
<br />
<br />
<br />
<br />
|-<br />
<br />
| | [[Image:FoldingSpeedDetection.png|Center|150px|]]<br />
| |<br />
<br />
== [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets|Shape Analysis With Overcomplete Wavelets]] ==<br />
<br />
In this work, we extend the Euclidean wavelets to the sphere. The resulting over-complete spherical wavelets are invariant to the rotation of the spherical image parameterization. We apply the over-complete spherical wavelet to cortical folding development [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> B.T.T. Yeo, W. Ou, P. Golland. On the Construction of Invertible Filter Banks on the 2-Sphere. Yeo, Ou and Golland. Accepted to the IEEE Transactions on Image Processing. <br />
<br />
P. Yu, B.T.T. Yeo, P.E. Grant, B. Fischl, P. Golland. Cortical Folding Development Study based on Over-Complete Spherical Wavelets. In Proceedings of MMBIA: IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis, 2007. [[Algorithm:MIT:ShapeAnalysisWithOvercompleteWavelets#Publication|More...]]<br />
<br />
<br />
|-<br />
<br />
| | [[Image:Brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:fMRI Clustering |fMRI clustering]] ==<br />
<br />
This type of algorithms assigns a tissue type to each voxel in the volume. Incorporating prior shape information biases the label assignment towards contiguous regions that are consistent with the shape model. [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration|More...]]<br />
<br />
<font color="red">'''New: '''</font> K.M. Pohl, J. Fisher, S. Bouix, M. Shenton, R. W. McCarley, W.E.L. Grimson, R. Kikinis, and W.M. Wells. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Accepted to the Special Issue of Best Selected Papers from MICCAI 06 in Medical Image Analysis [[Algorithm:MIT:Shape_Based_Segmentation_And_Registration#Publications|More...]]<br />
<br />
|-<br />
<br />
| | [[Image:MIT_DTI_JointSegReg_atlas3D.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_FiberRegistration|Joint Registration and Segmentation of DWI Fiber Tractography]] ==<br />
<br />
The goal of this work is to jointly register and cluster DWI fiber tracts obtained from a group of subjects. [[Algorithm:MIT:DTI_FiberRegistration|More...]]<br />
<br />
<font color="red">'''New:'''</font> U. Ziyan, M. R. Sabuncu, W. E. L. Grimson, Carl-Fredrik Westin. A Robust Algorithm for Fiber-Bundle Atlas Construction. MMBIA 2007<br />
<br />
<br />
|-<br />
<br />
| | [[Image:brain.png|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|Shape Based Level Segmentation]] ==<br />
<br />
<br />
This class of algorithms explicitly manipulates the representation of the object boundary to fit the strong gradients in the image, indicative of the object outline. Bias in the boundary evolution towards the likely shapes improves the robustness of the segmentation results when the intensity information alone is insufficient for boundary detection. [[Algorithm:MIT:Shape_Based_Level_Set_Segmentation|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:Wholebrain.jpg|thumb|left|200px]]<br />
| |<br />
<br />
== [[Algorithm:MIT:DTI_Clustering|DTI Fiber Clustering and Fiber-Based Analysis]] ==<br />
<br />
The goal of this project is to provide structural description of the white matter architecture as a partition into coherent fiber bundles and clusters, and to use these bundles for quantitative measurement. [[Algorithm:MIT:DTI_Clustering|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
Monica E. Lemmond, Lauren J. O'Donnell, Stephen Whalen, Alexandra J. Golby.<br />
Characterizing Diffusion Along White Matter Tracts Affected by Primary Brain Tumors.<br />
Accepted to HBM 2007.<br />
<br />
|-<br />
<br />
| | [[Image:Models.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Modeling|Fiber Tract Modeling, Clustering, and Quantitative Analysis]] ==<br />
<br />
The goal of this work is to model the shape of the fiber bundles and use this model discription in clustering and statistical analysis of fiber tracts. [[Algorithm:MIT:DTI_Modeling|More...]]<br />
<br />
<font color="red">'''New:'''</font> <br />
M. Maddah, W. M. Wells, S. K. Warfield, C.-F. Westin, and W. E. L. Grimson, Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts,IPMI 2007, Netherlands.<br />
<br />
|-<br />
<br />
| | [[Image:Thalamus_algo_outline.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_Segmentation|DTI-based Segmentation]] ==<br />
<br />
Unlike conventional MRI, DTI provides adequate contrast to segment the thalamic nuclei, which are gray matter structures. [[Algorithm:MIT:DTI_Segmentation|More...]]<br />
<br />
<font color="red">'''New:'''</font> Ulas Ziyan, David Tuch, Carl-Fredrik Westin. Segmentation of Thalamic Nuclei from DTI using Spectral Clustering. Accepted to MICCAI 2006.<br />
<br />
|-<br />
<br />
|-<br />
<br />
| | [[Image:ConnectivityMap.png|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:DTI_StochasticTractography|Stochastic Tractography]] ==<br />
<br />
This work calculates posterior distributions of white matter fiber tract parameters given diffusion observations in a DWI volume. [[Algorithm:MIT:DTI_StochasticTractography|More...]]<br />
<br />
<font color="red">'''New: '''</font><br />
<br />
|-<br />
<br />
| | [[Image:FMRIEvaluationchart.jpg|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:fMRI_Detection|fMRI Detection and Analysis]] ==<br />
<br />
We are exploring algorithms for improved fMRI detection and interpretation by incorporting spatial priors and anatomical information to guide the detection. [[Algorithm:MIT:fMRI_Detection|More...]]<br />
<br />
<font color="red">'''New:'''</font> Wanmei Ou, Sandy Wells, Polina Golland. Bridging Spatial Regularization And Anatomical Priors in fMRI Detection. In preparation for submission to IEEE TMI. <br />
<br />
|-<br />
<br />
| | [[Image:HippocampalShapeDifferences.gif|thumb|left|200px]]<br />
| | <br />
<br />
== [[Algorithm:MIT:Shape_Analisys|Population Analysis of Anatomical Variability]] ==<br />
<br />
Our goal is to develop mathematical approaches to modeling anatomical variability within and across populations using tools like local shape descriptors of specific regions of interest and global constellation descriptors of multiple ROI's. [[Algorithm:MIT:Shape_Analisys|More...]]<br />
<br />
<font color="red">'''New:'''</font> Mert R Sabuncu and Polina Golland. Structural Constellations for Population Analysis of Anatomical Variability.</div>Serdar