Centerline method CNN `sct_deepseg_sc` fails with certain subjects

ivadomed / model_seg_ms_mp2rage

Model repository for MS lesion segmentation on MP2RAGE data from University of Basel

MIT License

2 stars 0 forks source link

Centerline method CNN `sct_deepseg_sc` fails with certain subjects #10

Closed uzaymacar closed 2 years ago

uzaymacar commented 2 years ago

We initially called sct_deepseg_sc with -brain 1 -centerline cnn to speed up segmentation. However, a total of four subjects failed: sub-P004, sub-P011, sub-P006, sub-P017. In PR #7 with commit a869c68, I addressed this issue by changing the centerline method to be default option SVM which worked successfully on all subjects. I am opening this issue just in case there is more to be discussed on this subject.

uzaymacar commented 2 years ago

Closing this issue as we decided to go with the more robust SVM method for centerline extraction in preprocessing and:

We have done manual corrections on SC segmentations given with sct_deepseg_sc CNN method for centerline extraction.
Are working on a SC segmentation model (as part of the cascaded pipeline) which can further help with more robust SC segmentation.

jcohenadad commented 2 years ago

Should we get rid of the new option then? (for the sake of clarity) related to https://github.com/ivadomed/model_seg_ms_mp2rage/pull/17