neuropoly / data-management

Repo that deals with datalad aspects for internal use
4 stars 0 forks source link

`philadelphia-pediatric` - different conventions for disc labels #328

Open valosekj opened 5 days ago

valosekj commented 5 days ago

When working on the philadelphia-pediatric dataset in the context of https://github.com/ivadomed/model-spinal-rootlets/issues/19, @KaterinaKrejci231054 noticed that derivatives/labels contains disc labels with two different conventions.

Convention 1

Disc labels are located in the middle of the cord. For example, sub-101/anat/sub-101_rec-composed_T2w_labels-disc.nii.gz:

image ![sub-101](https://github.com/neuropoly/data-management/assets/39456460/dec16e7b-a411-44f4-b277-67a7d017663c)

Notice that the labels are located in the middle of the cord at the level of discs, i.e., not in the middle of each vertebral level as specified by our "Vertebral levels" convention.

Convention 2

Disc labels are located on the posterior tip of each disc. For example, sub-102/anat/sub-102_rec-composed_T2w_labels-disc.nii.gz:

image ![sub-102](https://github.com/neuropoly/data-management/assets/39456460/7e63471a-2fec-43e0-94ab-e3f9b50d7589)

When looking at the JSON sidecars, it seems that labels for different conventions were created by different authors:

valosek@macbook-pro:~/data/data.neuro.polymtl.ca/philadelphia-pediatric/derivatives/labels$ cat sub-101/anat/sub-101_rec-composed_T2w_labels-disc.json
{
    "Author": "Nadia Blostein",
    "Date": "2023-02-14 21:23:00",
    "Notes": "Corrected disc label 20, added disc label 21 (on SC, not on posterior disc)"
}
valosek@macbook-pro:~/data/data.neuro.polymtl.ca/philadelphia-pediatric/derivatives/labels$ cat sub-102/anat/sub-102_rec-composed_T2w_labels-disc.json
{
    "Author": "Sandrine Bedard",
    "Date": "2023-02-13 12:17:33"
}

Different disc positions (middle of the cord vs posterior tip of each disc) might have an impact on vert labeling (sct_label_vertebrae -discfile), registrations, or other processing steps. Should we unify the disc labels?

jcohenadad commented 5 days ago

Should we unify the disc labels?

yes, definitely. Tagging @NathanMolinier

NathanMolinier commented 5 days ago

Yes, this is the goal indeed to only follow our convention, however so far, the main method we have been using sct_label_vertebrae is outputting discs positions in the middle of the cord. This is the reason we have this distinction

valosekj commented 5 days ago

Thank you for the feedback!

Should we unify the disc labels?

yes, definitely.

Okay! We're gonna redo the discs to place them on the posterior tip of each disc.

NathanMolinier commented 5 days ago

If you want the discs labels to be projected onto the centerline, you can use the function sct_label_utils with the flag -project-centerline.

Screenshot 2024-06-25 at 13 01 20