Difference between revisions of "2010 Winter Project Week VervetMRILongitudinalAnalysis"

From NAMIC Wiki
Jump to: navigation, search
 
(2 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
__NOTOC__
 
__NOTOC__
  
<gallery perrow="5">
+
<gallery perrow="4">
 
Image:PW-SLC2010.png|[[2010_Winter_Project_Week#Projects|Projects List]]
 
Image:PW-SLC2010.png|[[2010_Winter_Project_Week#Projects|Projects List]]
 
Image:Calvin_atlas_demo.jpg | Axial slice for one of the vervet subjects
 
Image:Calvin_atlas_demo.jpg | Axial slice for one of the vervet subjects
Line 7: Line 7:
 
Image:Calvin_atlas_demo_vervet_gm.jpg | Aligned vervet grey matter probabilistic atlas
 
Image:Calvin_atlas_demo_vervet_gm.jpg | Aligned vervet grey matter probabilistic atlas
 
Image:Calvin_atlas_demo_segm.jpg | Overlayed segmentation result by EM Segmenter with vervet probabilistic atlas
 
Image:Calvin_atlas_demo_segm.jpg | Overlayed segmentation result by EM Segmenter with vervet probabilistic atlas
 +
Image:Calvin_ABC.jpg | Segmentation results by ABC
 +
Image:Calvin_EMSegment.jpg | Segmentation result by EMSegment in Slicer 2.8
 
</gallery>
 
</gallery>
  
Line 13: Line 15:
 
* Andriy Fedorov, Ron Kikinis (BWH)
 
* Andriy Fedorov, Ron Kikinis (BWH)
 
* Ginger Li, Chris Wyatt (VT)
 
* Ginger Li, Chris Wyatt (VT)
 +
* Martin Styner (UNC)
 +
* Kilian Pohl (IBM)
 +
* Hans Johnson (UIowa)
 +
* Marcel Prastawa (SCI)
  
  
Line 64: Line 70:
  
 
''During the meeting'':
 
''During the meeting'':
* testing and fine-tuning segmentation tools -- ''Work with: EM Segmenter/Kilian Pohl, itkEMS/ABC/Marcel Prastawa''; try to complete segmentation of baseline and followup scans
+
* discussed registration with Hans, Martin; looked at atlas to subject registration, confirmed accuracy is good for segmentation
* improving deformable registration of rhesus atlas to vervet, if needed ([[October_08_2009_Meeting:_VT-NAMIC_Revised_atlas_construction_workflow#Template_Gucci |initial results available]]) ''Work with: Luis Ibanez, Jim Miller, Dominik Maier, Casey Goodlett, Hans Johnson ... any expertise in registration is welcome''
+
* Worked with Marcel / ABC segmentation module -- worked out of the box, very few parameters need to be specified. Went through couple of iterations to expose some parameters of ABC (arbitrary number of classes, allow to disable registration, allow to specify priors per class).
* collect overall comments on the workflow from the related experience in similar workflows (Martin Styner)
+
* Worked with Kilian / EMSegmenter. Demo of Slicer2 EMSegmenter capabilities, not available in Slicer3. Fine-tuned parameters to achieve very good 3-class segmentation results on one subject.
 +
* Worked with Daniel / VMTK segmentation to remove vessels prior to segmentation.
 +
* Discussed the use of additional modalities for segmentation (T2 & PD). Agreed resolution not sufficient to improve results (4 times slice thickness). Single-shot T2/PD DICOM reading issue with Slicer nrrd reader: reported to Xiaodong
 +
 
  
 
</div>
 
</div>

Latest revision as of 17:34, 8 January 2010

Home < 2010 Winter Project Week VervetMRILongitudinalAnalysis



Key Investigators

  • Andriy Fedorov, Ron Kikinis (BWH)
  • Ginger Li, Chris Wyatt (VT)
  • Martin Styner (UNC)
  • Kilian Pohl (IBM)
  • Hans Johnson (UIowa)
  • Marcel Prastawa (SCI)


Objective

We are developing methods for analysis of structural images from nonhuman primates for identifying regional differences in brain morphometry caused by interactions of stress and alcohol exposure. Specifically, we are focused on vervet monkey species. Our objective is to develop image analysis workflow for accurate and reproducible segmentation of vervet brain to facilitate morphometry studies. Ideally, we would like to have the complete workflow implemented using 3DSlicer tools.

Approach, Plan

Given two serial T1 scans for 10 vervet subjects, our approach is the following:

1. (Semi-)manually segment GM/WM/CSF and certain subcortical structures for one of the subjects

2. Build probabilistic tissue atlas using the baseline images

3. Register follow-up to baseline

4. Use the atlas and automated segmentation tools to identify subcortical structures in all subjects for the baseline and follow-up

Alternatively, instead of the Steps 1 and 2 we could use UNC Rhesus atlas, if the structural changes between the rhesus and vervet brains are not significant, and we can successfully perform deformable registration.

The focus of our work at AHM will be on Step 4 above. Given two types of atlases:

  • Vervet atlas we constructed
  • Rhesus atlas prepared by UNC group

... and three approaches to segmentation:

  • EM Segmenter in Slicer 3.4
  • itkEMS and/or ABC Slicer 3.5 extension (are they different?)
  • fluid deformable registration of the atlas to a subject

we will study what combination works best for our data, based on the subjective evaluation of the segmentation results.

Progress

Prior to the meeting (current progress and notes):

  • completely redesigned atlas construction workflow, prepared detailed documentation
  • prepared 3-class and subcortical segmentations for one subject (semi-manually)
  • atlas construction workflow is now composed completely from Slicer/NA-MIC Kit tools
  • initial version of the re-worked vervet probabilistic atlas completed

During the meeting:

  • discussed registration with Hans, Martin; looked at atlas to subject registration, confirmed accuracy is good for segmentation
  • Worked with Marcel / ABC segmentation module -- worked out of the box, very few parameters need to be specified. Went through couple of iterations to expose some parameters of ABC (arbitrary number of classes, allow to disable registration, allow to specify priors per class).
  • Worked with Kilian / EMSegmenter. Demo of Slicer2 EMSegmenter capabilities, not available in Slicer3. Fine-tuned parameters to achieve very good 3-class segmentation results on one subject.
  • Worked with Daniel / VMTK segmentation to remove vessels prior to segmentation.
  • Discussed the use of additional modalities for segmentation (T2 & PD). Agreed resolution not sufficient to improve results (4 times slice thickness). Single-shot T2/PD DICOM reading issue with Slicer nrrd reader: reported to Xiaodong



References