philadelphia-pediatric: organization and naming of new subjects

[valosekj]

This issue discusses data organization and file naming for new subjects for the philadelphia-pediatric dataset that our collaborators will share with us.

Current organization and naming:

├── anat
│   ├── sub-172_T1wSlab1.json
│   ├── sub-172_T1wSlab1.nii.gz
│   ├── sub-172_T1wSlab2.json
│   ├── sub-172_T1wSlab2.nii.gz
│   ├── sub-172_T2wSlab1.json
│   ├── sub-172_T2wSlab1.nii.gz
│   ├── sub-172_T2wSlab2.json
│   └── sub-172_T2wSlab2.nii.gz
├── DWI
│   ├── sub-172_DWI.bval
│   ├── sub-172_DWI.bvec
│   ├── sub-172_DWI.json
│   └── sub-172_DWI.nii.gz
└── composed
    ├── sub-172_T1COMP.json
    ├── sub-172_T1COMP.nii.gz
    ├── sub-172_T2COMP.json
    └── sub-172_T2COMP.nii.gz

Organization and naming that would be compatible with our convention:

├── anat
│   ├── sub-172_acq-bottom_run-1_T1w.json
│   ├── sub-172_acq-bottom_run-1_T1w.nii.gz
│   ├── sub-172_acq-bottom_run-1_T2w.json
│   ├── sub-172_acq-bottom_run-1_T2w.nii.gz
│   ├── sub-172_acq-top_run-1_T1w.json
│   ├── sub-172_acq-top_run-1_T1w.nii.gz
│   ├── sub-172_acq-top_run-1_T2w.json
│   ├── sub-172_acq-top_run-1_T2w.nii.gz
│   ├── sub-172_rec-composed_T1w.json
│   ├── sub-172_rec-composed_T1w.nii.gz
│   ├── sub-172_rec-composed_T2w.json
│   └── sub-172_rec-composed_T2w.nii.gz
└── dwi
    ├── sub-172_run-1_dwi.bval
    ├── sub-172_run-1_dwi.bvec
    ├── sub-172_run-1_dwi.json
    └── sub-172_run-1_dwi.nii.gz

Changes:

• directory renaming: DWI to dwi
• file renaming: usage of acq-bottom and acq-top keys in the file names. Usage of rec-composed for composed images. Putting composed images into the anat directory.

[valosekj]

@jcohenadad, I'm now thinking whether run-1 is actually necessary. We use it for the subjects we already have. But we never use run-2:

cd philadelphia-pediatric
$ ls sub-1*/anat/*run-2*.nii.gz
ls: sub-1*/anat/*run-2*.nii.gz: No such file or directory

But for consistency with the already shared data, I would use run-1 also for new subjects.

[jcohenadad]

I think that top/bottom should use the chunk entity instead. Example: sub-172_acq-bottom_run-1_T1w.json should be replaced with sub-172_chunk-2_run-1_T1w.json

[jcohenadad]

@jcohenadad, I'm now thinking whether run-1 is actually necessary. We use it for the subjects we already have. But we never use run-2:

If we don't have run-2, then I agree, there is no need for run-1

[valosekj]

Agree with both. Let's share the data with filenames proposed in the first comment. And I will then fix the filename for the whole dataset.

[valosekj]

During a BIDS validation of the newly shared data, @mguaypaq discovered that several subjects have empty sform_code and qform_code, leading to unknown image orientation. Details can be found at this data.neuro.polymtl.ca comment.
When opened in FSLeyes, these images display red question marks instead of R/L, A/P, S/I labels:

screenshot

I contacted our collaborators about this issue and received the following response:

Regarding the spatial orientation, I believe Nadia Blostein used the data in that same format without correction on our end for template creation. If it’s possible to check with her she may already have a solution for the orientation issues which works with your pipeline. We’ve used a couple of methods to fix the orientations but just want to be sure that the solution works for you.

I've looked in our repos, but I haven't found any piece of code related to the orientation fix. @jcohenadad, @rohanbanerjee, @NadiaBlostein, do you happen to have any insights?

[valosekj]

Follow-up on the empty sform_code and qform_code from my previous comment. After iteratively promoting Claude while controlling its output in FSleyes, I ended up with the following commands to fix the image header:

# creaty a copy image to play with
cp sub-127_rec-composed_T2w_copy.nii.gz sub-127_rec-composed_T2w_fixed.nii.gz
# fix qform
fslorient -setqformcode 1 sub-127_rec-composed_T2w_fixed.nii.gz
fslorient -setqform 0 0 -1 0 -1 0 0 0 0 1 0 0 0 0 0 1 sub-127_rec-composed_T2w_fixed.nii.gz
# use the same transformation for sform:
fslorient -setsformcode 1 sub-127_rec-composed_T2w_fixed.nii.gz
fslorient -setsform 0 0 -1 0 -1 0 0 0 0 1 0 0 0 0 0 1 sub-127_rec-composed_T2w_fixed.nii.gz
# now, the image is in AIL
#  --> reorient it to LPI to be in the same orientation as sub-127_acq-top_run-1_T2w.nii.gz to allow comparison in FSLeyes
sct_image -i sub-127_rec-composed_T2w_fixed.nii.gz -setorient LPI -o sub-127_rec-composed_T2w_fixed_LPI.nii.gz

Now, when opening sub-127_acq-top_run-1_T2w.nii.gz and sub-127_rec-composed_T2w_fixed_LPI.nii.gz in two separate FSLeyes windows, I can compare the axes. S-I and A-P are easy. R-L vs L-R is tricky (as the cord is symmetric); I used a slight scoliosis to match R-L between images.
However, when double-checking some already existing images, for example, for sub-123, I noticed that the R-L axis is swapped! --> Opening a separate issue: #337

[valosekj]

Update: I received fixed composed images from collaborators and pushed them to data.neuro.polymtl.ca; details in https://data.neuro.polymtl.ca/datasets/philadelphia-pediatric/pulls/3#issuecomment-624.