Update spinal cord seg of T2w with contrast-agnostic v3.1

[sandrinebedard]

Description

This PR updates spinal cord segmentations of T2w images with the contrast-agnostic 3.1 version. Visual QC was done to assess the quality of the segmentations, but no manual correction was done.

It only adds the segmentations for subject that don't have mild compression, n=203 (as marked in participants.tsv).

These segmentations were used to computed normative values in spinalcordtoolbox/PAM50-normalized-metrics#20

We should create another release of the dataset when this PR is merged.

[valosekj]

Thanks @sandrinebedard, LGTM!

It only adds the segmentations for subject that don't have mild compression, n=203 (as marked in participants.tsv).

Yeah, so this means that the other subjects (those with mild compression) still have cord segmented using sct_deepseg_sc. But as the segmentation method is tracked using JSON sidecars and all segmentations were QCed anyway, I hope it's okay.

[sandrinebedard]

Yeah, so this means that the other subjects (those with mild compression) still have cord segmented using sct_deepseg_sc. But as the segmentation method is tracked using JSON sidecars and all segmentations were QCed anyway, I hope it's okay.

Yes! As discussed in person with @jcohenadad, since the segmentations done with sct_deepseg_sc were manually corrected, so we know the segmentation is accurate at the compression site, compared to the contrast-agnostic inference, it would be counterproductive to overwrite them!

[mguaypaq]

Git-annex looks fine, and bids-validator is irrelevant for derivatives/.

I noticed that some JSON files had duplicate lines, so I removed the duplicates:

    "SpatialReference": "orig",
    "SpatialReference": "orig",