branch 3.6 gives improper and lower quality segmentations in specific tracks

[bacaron]

Hey,

I've had previous runs of the segmentation app on branch 3.3 that generated overall great segmentations in most tracks, especially the uncinate and ILF (see screenshots). However, I have a few subjects missing ILFs, prompting need for further investigation and rerunning of the app.

I recently ran the latest version of the app this morning (branch 3.6), and although it returns both ILFs in a subject that had previously been missing them, it has now caused a reduction in quality of the ILF from the segmentation generated from branch 3.3 and the uncinate (see screenshots).

Comparing change logs, it appears there have been multiple updates to wma_tools but it's hard to gather what the specific changes in code are and how they would relate to the differences in segmentation.

Is it possible some old code might have snuck its way into the newest branch? Or did the new changes implemented since branch 3.3 lead to the observed differences in segmentations.

Happy to provide the data that generated both segmentations, and the segmentations themselves.

Thanks,
Brad

Branch 3.3 right uncinate: good

branch 3.6 right uncinate: not good

branch 3.3 right ilf: good

branch 3.6 right ilf: not good

[soichih]

@bacaron are you still seeing this problem on the latest version (3.8)?