Hi Spharm experts, 

I'm running a SPHARM analysis of roughly 200 subjects via the command line and I'm finding that some participants (roughly 10%)  stalled out during the GenParaMeshCLP step. Any idea on why this could be? I've double checked the raw binary input (.nii.gz) as well as the .gipl file created by SegPostProcessCLP, and both look good.

Advice on how to troubleshoot this problem would be much appreciated.

Thanks!
Lara

Hi Lara,

I've had the GenParaMeshCLP hang where it doesn't seem to be able to get past the first "step" - i.e. there isn't the usual multitudes of iterations output into the terminal window.

If it's the same error you're getting, I think it's due to shapes which are slightly too "pointy" or irregular for spharm to handle. I've fixed it by either editing back the original segmentation a little, or using a little bit of Gaussian smoothing (I start with 0.5 or 1) and increment the smoothing if necessary.

Hope this helps,
Dave.

Thanks Dave
I agree that's what we are doing too, usually I first edit manually the segs by adding/removing voxels such that very pointy extrusions and protrusions are reduced. In order to change the segmentation as little as needed, I usually do a few voxels and then rerun to see whether that resolved it.

At issue is basically the initial mapping to the sphere using the heat-equation mapping, which can create a "bunching-up" of triangles. We are working on improving that initial mapping such that this won't be necessary anymore.

Martin

Hi Dave and Lara,



I wanted to add to what Dave said sometimes a closed c-shaped geometry ("almost hole") will create the stalling in GenParaMesh. SegPostProcess takes care of most of those things, but sometimes in very complicated geometries, the problem remains after SegPostProcess. What Martin suggests is right, we do minimally and manually edit the segmentations to avoid the problem. 


Lara, please let us know if that works for you.
Thanks,


Beatriz

Great. Thank you for all of these suggestions! I will re-examine my binary roi mask inputs and see if minor edits eliminate the problem.

Thanks again!
Lara