Add segmentations from William Sirois

spine-generic / data-multi-subject

Multi-subject data for the Spine Generic project

Creative Commons Attribution 4.0 International

22 stars 15 forks source link

Add segmentations from William Sirois #172

Open mguaypaq opened 1 month ago

mguaypaq commented 1 month ago

These are the files provided by William Sirois at the end of his GBM3100 internship. Todo:

Figure out which derivatives/ subfolder they should go into.
Rename the files according to BIDS.
Add JSON sidecars.

While uploading the files, I noticed that most of them are new, but a few of them are exact duplicates of already existing files in the repository.

sandrinebedard commented 1 month ago

I added to the branch the spinal cord canal seg under derivatives/labels, I am not sure that it worked, @mguaypaq can you check?

mguaypaq commented 1 month ago

It looks like this is what happened:

1 of the 188 images from William Sirois (sub-barcelona02_011) got replaced by a new image from Sandrine (sub-barcelona02_space-other_T2w_label-canal_seg.nii.gz).
4 of the 188 images from William Sirois were exact copies of existing canal_seg files under derivatives/labels/.
6 of the 188 images from William Sirois replaced existing canal_seg files under derivatives/labels/.
The remaining 177 of 188 images from William Sirois are new canal_seg files under derivatives/labels/.

But:

There are also 80 new sub-*_T2w_label-discs_dlabel.nii.gz images from Sandrine under derivatives/labels/.

@sandrinebedard is this what you wanted?

sandrinebedard commented 1 month ago

But:

There are also 80 new sub-*_T2w_label-discs_dlabel.nii.gz images from Sandrine under derivatives/labels/.

@sandrinebedard is this what you wanted?

Nope hahaha, just realized a script I was running was modifying the files... sorry, we should remove those changes

mguaypaq commented 1 month ago

Ok, I can remove the extra modifications. Do we also need to create and/or update any JSON sidecars with the provenance of the data?

sandrinebedard commented 1 month ago

Abel just created them yesterday, I will push them today!

mguaypaq commented 3 weeks ago

I don't think these are the right JSON sidecars; they describe the 267 images mentioned at the end of https://github.com/ivadomed/model-canal-seg/issues/7, but not the 188 images mentioned at the start (which are the ones in this PR). We probably need at least 4 different JSON files:

The 4 images that were already in this dataset before the PR have existing JSON sidecars, so we should use them.
The 1 image that was modified in https://github.com/spine-generic/data-multi-subject/pull/172/commits/5c22e3d83da3a9d5cca4cf83e77a511b273a2a4d (instead of being just moved) needs an explanation of what was done to modify it.
The 6 images that replace existing canal segs, we should check whether they are manual corrections, or the result of an earlier model, or what.
The 177 other images, we should figure out how they were produced.

That's if we want the 188 images used as ground truth for the training. Alternatively, if we want the 267 images produced by the final model, we could upload them instead? Or, if we're not sure of the origin, we should say that in the JSON files.