Failed segmentation

Dear all,

I am trying to test SCT in a MacOS 13.1, I am trying to test the segmentation tool (sct_deepseg_sc) with the attached .nii file, but I am unable to get results. I hope you can help sort if this is because if the image acquisition or is it something with my SCT?

<sct_deepseg_sc -i m21.nii.gz -c t1 -qc qc>

(I am unable to upload the file because is ~9 mb, I am attaching a link to my drive:
Microsoft OneDrive - Access files anywhere. Create docs with free Office Online.)

Please help!

Hi @enriquegf,

Thank you for reaching out. Indeed, I confirm the segmentation fails. Looking deeper (by using the flag -r 0 to keep intermediate outputs), I see that the centerline detection failed (yellow curve):

When using the flag -centerline viewer, you can create a reliable centerline as illustrated here:

When using that centerline, the segmentation is improved. However, it is far from being perfect, mostly because the acquisition is not prone to segmenting the spinal cord reliably (ie: sagittal acquisition with thick slices, in your case 4.4mm). So, I tried with 3D kernel, and also with PropSeg. The latest gave the best results. See comparison (red: sct_deepseg_sc -kernel 2d, yellow: sct_deepseg_sc -kernel 3d, green: sct_propseg):


See QC with syntaxes used: (3.3 MB)

And the centerline so you can reproduce my results: m21_centerline.nii.gz (37.9 KB)