Bridge clipping incomplete about cobiveco HOT 5 OPEN

Looking a little closer, it seems like the distance threshold set on line 108 is not actually used in the subsequent point selection using rangesearch a few lines down. Instead, a fixed value of 1.0 is used. Using the actual threshold value set as a 15th percentile seems to always be too small (less than the default 1). It seem like this radius should be an exposed parameter or dynamically set and increased until a proper clipping of the bridge is achieved.

After playing around a bit here is what I found that worked.
I changed the threshold to be as follows. basing it off the max of dist2 plust some fudge factor makes it more likely that the points identified on the epi surface will actually form a closed path to isolate the bridge.

107 % use tenths percentile as threshold for distance search
108  threshold = max(dist2)+prctile(dist2,75);

I then changed the code that identifies the nodes of the path to actually use this threshold

[idx1,D1] = rangesearch(mesh.vol.points,pointsInPlane1Filtered,threshold,'NSMethod','exhaustive');

[idx2,D2] = rangesearch(mesh.vol.points,pointsInPlane2Filtered,threshold,'NSMethod','exhaustive');

from cobiveco.

Looping in @lisapankewitz as this is code added in CobivecoX

Congrats on your recently accepted paper! https://doi.org/10.1016/j.media.2024.103091
If there are any related code changes, we'd be happy to see them synchronized so that the description in the paper and the code here match.

from cobiveco.

Dear Jake,

thanks so much for this feedback!
I agree, this should be an exposed parameter!

This issue can be a bit tricky sometimes, due to the local curvature of the inflow outflow tracts especially in certain cases on the LV, I found.

I appreciate sharing your findings. I will include this in my next pull request!

@axel-loewe Thanks so much!!! I am submitting a major Pull request this week, once I have the feedback of the proof. Just wanted to avoid you going through several major PRs. It will reflect the code changes and a few additional documentation changes.

from cobiveco.

I am having this same issue again on a new mesh, and I am wondering if it could be resolved in how the epi path is being defined. In the example below, I could up the threshold for finding the cut points, but it would be more robust if the epi points actually followed the whole path. I implemented a change where the epi path is calculated the same way the endo path is, just using an epicardial only graph. The results seem to work ok, and it also means that the distance threshold for finding nearby points for clipping can be turned way down. Changes will be pushed onto my fork once I clean them up a bit.


After changes:

from cobiveco.

Hi Jake,
I originally (before pushing the code here), had the paths on Epi and endo created in the same way. However, especially in the congenital heart disease cases, I had issues excising the bridges correctly if defining the path on the epi the way I do it in the endo path (changes is local curvature led to different paths solutions for epi and endo paths). So I think, this solution can work, but depends on a few factors including changes in wall thickness, etc.

from cobiveco.