[ITK] Geodesic active contour liver segmentation in 3D

[Jonathan Young]

Hello all,

I am trying to reconstruct a 3D model of the liver from an MRI of a patient
with hepatocellular carcinoma. I have been able to obtain satisfactory
segmentations using the geodesic active contour method in SimpleITK, but
only when operating on a single slice at a time. When I tried extending the
method to 3D by working with a series of image slices at once, the
segmentation seems to fail. The initial level set propagates beyond the
boundaries of the liver in a few slices, while in another slice it doesn't
move at all. I did not have these issues in the 2D case.

My work is done in an IPython notebook which can be found here:

https://nbviewer.jupyter.org/github/jon-young/medicalimage/blob/master/Liver%20Segmentation%203D.ipynb

At the bottom of the notebook, displaying the result of the segmentation
shows the problems I described above. I don't think it is a matter of the
number of iterations being run, as with fewer iterations the same issues
occur. Could the problem stem from the way I've setup the initial level
set? If anyone could take a look to see what I've done wrong, I would
greatly appreciate it.

Thanks,
Jonathan

-- 
Jonathan Young
Graduate Research Assistant, Laboratory of Edward M. Marcotte
Center for Systems and Synthetic Biology
The University of Texas at Austin
2500 Speedway, MBB 3.128, Austin, TX 78712
(512) 232-3919
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20151209/04cf361c/attachment.html>