Difference between revisions of "2011 Winter Project Week:RegistrationAnisotropy"

From NAMIC Wiki
Jump to: navigation, search
 
(10 intermediate revisions by 2 users not shown)
Line 1: Line 1:
 
__NOTOC__
 
__NOTOC__
<!--
+
<gallery widths="400px" perrow="2">
[[Image:RegEval_ImageHistory_3mm.png|400px|Example image processing pipeline]] Example processing pipeline simulating PV effects from anisotropic voxel size <br>
+
Image:RegEval_JointHist_AGif.gif|Fig.1: Example joint histogram blurring effects from anisotropic voxel size : 1x to 20x anisotropy
[[Image:RegEval_JointHist_3mm.jpg|400px|Example joint histogram]] Example joint histogram blurring effects from anisotropic voxel size: '''leftmost''': reference blurring from interpolation blurring alone (no PV); middle: 3x1 anisotropy against original; '''right''': 3x1 anisotropy against blurred reference <br>
+
Image:RegEval_JHBlur_RotX_AGif.gif|Fig.2: for comparison: example joint histogram blurring effects from rotation around x (LR) axis 1-15 degrees
-->
+
Image:RegEval_JHBlur_RotZ_AGif.gif|Fig.3: for comparison: example joint histogram blurring effects from rotation around z (IS) axis 1-15 degrees
<!--
+
Image:MMI_plot.png|Fig.4: MMI optimization landscape (1-DOF) blurred (red) by anisotropy. We measure the width of the blurred peak or relate the drop in intensity to an equivalent shift away from the optimum.
[[Image:RegEval_JointHist_5mm.jpg|400px|Example joint histogram]] Example joint histogram blurring effects from anisotropic voxel size: '''leftmost''': reference blurring from interpolation blurring alone (no PV); middle: 3x1 anisotropy against original; '''right''': 5x1 anisotropy against blurred reference
 
-->
 
  
=The 3DSlicer Registration Case Library Project=
+
</gallery>
 +
 
 +
 
 +
= Voxel Anisotropy in Registration =
  
 
==Key Investigators==
 
==Key Investigators==
* BWH: Dominik Meier,  Andryi Fedorov
+
* BWH: Dominik Meier,  Andriy Fedorov, William Wells
  
 
<div style="margin: 20px;">
 
<div style="margin: 20px;">
Line 27: Line 28:
  
 
Aside from designing experiments for metrics that are specific to this effect, we also seek interaction with other projects that have highly anisotropic image data.   
 
Aside from designing experiments for metrics that are specific to this effect, we also seek interaction with other projects that have highly anisotropic image data.   
 +
Example Experiment: produce directly joint histograms and difference(ratio) images w/o the need to run the actual registration, i.e. we compare the effects of voxel anisotropy on the theoretical optimum: Move, filter and subsample identical image pair, then resample back to original position and build joint histograms and subtraction images. Because of the increasing PV effects we expect to see a degenerating joint histogram and a subtraction image with increasing edge artifacts. We can then try to interpret how the optimizer will behave in this environment. The benefit of this metric is that we circumvent the stochastic nature of the registration output.
 
<br>
 
<br>
  
Line 34: Line 36:
  
 
<h3>Progress</h3>
 
<h3>Progress</h3>
*see below: [[2011_Winter_Project_Week:RegistrationAnisotropy#Progress]]
+
*see images above
 +
*we identified '''prostate MRI''' as a promising dataset for testing. Since hi-res isotropic data is never avail., we will move from the current anisotropy (x7) toward isotropic. Also we have 2 orientations (coronal vs. sagittal) and can compare the effects of having the anisotropy run in a different direction.
 +
*progress on exposing the '''ITK joint histogram''' functions and MI metric for quantifying the above shown effects.
 +
*we anticipate the optimum in the MI landscape to be blurred and potentially shifted by anisotropy. Using translation as a 1-DOF experiment, we calculate the new landscape and measure the blurring (and ev. shift) of the peak as a metric for reduction in accuracy and robustness
 
</div>
 
</div>
 
</div>
 
</div>
  
 
<div style="width: 97%; float: left;">
 
<div style="width: 97%; float: left;">
 +
 
== Progress ==
 
== Progress ==
  project links go here
+
*see above
  
  
==References==
 
*[[2011_Winter_Project_Week#Projects|back to AHM_2011 project list]]
 
*[[Projects:RegistrationDocumentation|Link to the RegistrationDocumentation Project Wiki]]
 
*[[Projects:RegistrationDocumentation:RegLibTable|Link to Registration Case Library]]
 
  
 
</div>
 
</div>

Latest revision as of 17:29, 14 January 2011

Home < 2011 Winter Project Week:RegistrationAnisotropy


Voxel Anisotropy in Registration

Key Investigators

  • BWH: Dominik Meier, Andriy Fedorov, William Wells

Objective

For this work we seek insight into the effects of voxel anisotropy and image inhomogeneity on registration accuracy and robustness.


Approach, Plan

Aside from designing experiments for metrics that are specific to this effect, we also seek interaction with other projects that have highly anisotropic image data. Example Experiment: produce directly joint histograms and difference(ratio) images w/o the need to run the actual registration, i.e. we compare the effects of voxel anisotropy on the theoretical optimum: Move, filter and subsample identical image pair, then resample back to original position and build joint histograms and subtraction images. Because of the increasing PV effects we expect to see a degenerating joint histogram and a subtraction image with increasing edge artifacts. We can then try to interpret how the optimizer will behave in this environment. The benefit of this metric is that we circumvent the stochastic nature of the registration output.

Progress

  • see images above
  • we identified prostate MRI as a promising dataset for testing. Since hi-res isotropic data is never avail., we will move from the current anisotropy (x7) toward isotropic. Also we have 2 orientations (coronal vs. sagittal) and can compare the effects of having the anisotropy run in a different direction.
  • progress on exposing the ITK joint histogram functions and MI metric for quantifying the above shown effects.
  • we anticipate the optimum in the MI landscape to be blurred and potentially shifted by anisotropy. Using translation as a 1-DOF experiment, we calculate the new landscape and measure the blurring (and ev. shift) of the peak as a metric for reduction in accuracy and robustness

Progress

*see above