Deepseg missing the mark?

HI SCT team,

I wasn’t able to reply to the existing ticket on sourceforge, so will carry on here instead :slight_smile:

The plot thickens…please bear in mind that I’ve always used the viewer to supply the center line of the spinal cord, but just for kicks I tried running deepseg without the viewer, instead letting it use the default svm method.

sct_deepseg_sc -i sanlm_structural.nii.gz -c t1 -ofolder deepseg

which I wanted to compare to my standard call:

sct_deepseg_sc -i sanlm_structural.nii.gz -c t1 -ofolder deepseg.viewer -centerline viewer

In short, selecting the centre of the cord (really really carefully) makes deepseg fail. See attached.

RED = deepseg automatic, BLUE = deepseg with viewer selected center line (+ overlap for comparison). (Please note: I’ll have to post the images in separate replies)

Hope this helps - I will send Julien a link to the data set - but please note this is not specific to just this subject - I see this behaviour consistently.

Best wishes,


Thank you very much for this bug report Jon!
We have opened a ticket which will be addressed ASAP.