Update spinal levels based on Frostell et al. paper

[jcohenadad]

This PR updates the spinal levels with an improved version. See full context in #16.

Generated with: d854c26

As the new spinal levels are integer (vs. the previous spinal levels which were probabilistic), they are contained within a single file located under PAM50/template (previously PAM50/spinal_levels). The following changes are therefore necessary:

• On the PAM50 side, fix the spinal cord segmentation The spinal cord in PAM50_cord.nii.gz is missing the caudal part #19
• Generate PAM50_spinal_levels.nii.gz under PAM50/template
• Start a discussion on the forum to get feedback from researchers about this plan
• Remove PAM50/spinal_levels
• On the SCT side: when running sct_warp_template, remove the option -s because the spinal levels will be warped by default alongside all files under PAM50/template
• Regenerate 14 (point-labels) from the new full-cord level labels file
• Add correct entry (15) in info_label.txt
• Update documentation for new spinal levels spinalcordtoolbox#4247
  • PR: Update documentation for new spinal levels spinalcordtoolbox#4249
• Update SCT course for new spinal levels spinalcordtoolbox#4248
  • Unresolved course comment

Fixes #16

Related: #21

[jcohenadad]

As per #28 (comment), a suggestion would be to have:

• one file for spinal level labels (one pixel per level): 14, point-wise spinal level labels, PAM50_label_spinal_levels.nii.gz
• one file for spinal levels (labeled segmentation): 15, spinal cord mask with spinal levels, PAM50_spinal_levels.nii.gz

[jcohenadad]

Oops! on 38f6ce3 I noticed a value=1 at the very bottom:

It extends for 9 slices. Probably related to the code, which rounds segment length and then accumulate errors going caudally:

PAM50/scripts/generate_spinal_levels.py

Line 94 in 38f6ce3

length_segment = np.uint8(length_spinalcord * level_percent / 100)

[jcohenadad]

We can merge and fix spinalcordtoolbox/spinalcordtoolbox#4248 afterwards