search

ivadomed / model-spinal-rootlets

Deep-learning based segmentation of the spinal nerve rootlets
5 stars 2 forks source link

Artifact in hc-leipzig-7t-mp2rage dataset in C2 level #69

Open KaterinaKrejci231054 opened 3 months ago

KaterinaKrejci231054 commented 3 months ago

Artifact in hc-leipzig-7t-mp2rage dataset in C2 level

During my manual corrections of the rootlets segmentations, I found that almost all data contain an artifact in the C2 region (see below) on the R side. Because of this, the rootlets are not visible in this region. This may be the reason why the nnUNet model is unable to train in this region.

Point for discussion: Do we require manual corrections in the C2 region to train the model? Or will we consider the C2 level as "empty" in this case?

image

jcohenadad commented 3 months ago

Do we require manual corrections in the C2 region to train the model? Or will we consider the C2 level as "empty" in this case?

I would say, whenever we see the rootlets, we should segment it

valosekj commented 3 months ago

Interesting finding! Given that a similar artifact (C2, right side) appears across multiple subjects, might there be some systematic problem during the data acquisition?