Difference between revisions of "2013 Project Week:4DShapeAnalysis"

From NAMIC Wiki
Jump to: navigation, search
Line 30: Line 30:
 
<h3>Progress</h3>
 
<h3>Progress</h3>
  
 +
'''Segmentation'''
 +
 +
Here we consider a multi-shape complex of caudate, putamen, and thalamus, which were provided as binary images.  Triangular meshes were extracted using marching cubes, giving us 3 shapes at 3 time points for each subject.
 +
 +
'''Regression'''
 +
 +
The continuous evolution of all 3 structures is estimated jointly
 +
 +
'''Conclusion'''
 +
 +
* As before, segmentation quality is the most important factor in the analysis.  The changes we are interested in detecting could potentially be of the same order as segmentation error.
 +
* Estimating the baseline shape may improve the robustness of the estimation,
 +
* However, we need to be careful about the estimation of the baseline shape.  The changes over time are very subtle, therefore small differences in the baseline shape could lead to completely different dynamics of shape change. 
 
</div>
 
</div>
 
</div>
 
</div>

Revision as of 16:14, 11 January 2013

Home < 2013 Project Week:4DShapeAnalysis

Key Investigators

  • James Fishbaugh, Marcel Prastawa, Guido Gerig, SCI
  • Hans Johnson, Iowa

Project Description

Objective

In past project weeks, we have explored the application of shape regression to study sub-cortical changes associated with Huntington's disease. This has led to the development and release of open source shape analysis software. During this project week, we would like to investigate a new algorithm recently developed for our software. The method is the extension of linear regression to shapes, where we jointly estimate a baseline shape (intercept) and initial momenta vectors (slope) which define the dynamics of shape change over time. This is in contrast to the current implementation, which assumes the baseline shape is fixed at the earliest time point.

Approach, Plan

  • Individual regression with control points at shared locations
    • Compare to previous results
    • Compare differences between HD subjects and controls

Progress

Segmentation

Here we consider a multi-shape complex of caudate, putamen, and thalamus, which were provided as binary images. Triangular meshes were extracted using marching cubes, giving us 3 shapes at 3 time points for each subject.

Regression

The continuous evolution of all 3 structures is estimated jointly

Conclusion

  • As before, segmentation quality is the most important factor in the analysis. The changes we are interested in detecting could potentially be of the same order as segmentation error.
  • Estimating the baseline shape may improve the robustness of the estimation,
  • However, we need to be careful about the estimation of the baseline shape. The changes over time are very subtle, therefore small differences in the baseline shape could lead to completely different dynamics of shape change.

Introduction

HD Subjects

HD


Controls