[ITK] Troubles using level-sets methods

[Timothee Evain]

Hello ITK community,

First of all, this is my first post to this list, whereas I'm using ITK since some years, and I would like to say that I think it's a really great toolkit, so many thanks to developers! Keep going.

What pushs me to ask for help is level-set segmentation:

I'm currently working on maxillo facial 3D CBCT data, tying to segment vessel-like dental structures with a GeodesicActiveContourLevelSetImageFilter.
My inputs are an initial level set distance map obtained with SignedDanielsson filter coming from a coarse binary pre-segmentation, with is quite representative of my structure, but not enough.
The feature image is a potential egde map as advised in the Software guide, i.e. a gradient magnitude of smoothed data put through a Sigmoid filter. From what I understood by looking for older answers, this map is more or less the critical element to obtain good segmentation. As much as I can tell, in my case it seems to be good with the K1/K2 rule of thumb for tuning the Alpha and Beta parameter.

But here is the problem: when I run the filter, the result is almost the same as the initial input, whatever Advection/Propagation/Curvature scaling parameter is set. I could run for 1 or 100 iterations, I see no evolution of contour (even if i just put Propagation force).
Since it is quite difficult to setup an efficient evolving visualization for 3D, I can't tell want happens here.

I tried with the ITK Snap method and succeded to get better results, but for what I know, the level set here is lightly modified.
Any idea on what is wrong ? Can my data be too complicated for this levelset method ?


Tim

P.S. : I could post the code if anyone want, but it is almost similar to the example one given in the SoftwareGuide