2012 Winter Project Week:EndoSeg

• 

  Projects List

• 

  Atlas Image

• 

  Non-atlas Image

• 

  Axial view of segmentation result. Red contour: manual segmentation; Green mask: automatic segmentation

• 

  Sagittal view

• 

  Coronal view

Key Investigators

Yi Gao, Liang-Jia Zhu, Josh Cates, Alan Morris, Danny Perry, Greg Gardner, Rob MacLeod, Sylvain Bouix, Allen Tannenbaum

Objective

• We are developing methods for segmenting the endocardium from the DE-MRI. The current proposed method is a multi-atlas based registration approach. The user is presented with a weighted average of all registered atlas segmentations. The final LA segmentation is a thresholded volume from this weighted average. Weighting is based on the accuracy of the registration, as determined by a mutual information metric.
• There is a prototype module developed by Yi that is currently in Slicer. We propose to expand and tweak this module to specifications determined by our experience using the algorithm on many real patient datasets.
• We will also discuss posssible collaborative paper ideas based on this work.

Approach, Plan

• Improve the speed and accuracy using the multi-scale scheme
• Improve the accuracy of atlas segmentation accuracy with better label fusion techniques.

Progress

The current segmentation module is in the Slicer extension manager.

Delivery Mechanism

This work will be delivered to the NA-MIC Kit as a (please select the appropriate options by noting YES against them below)

1. Slicer Module
   1. Extension -- commandline

References