Quality problem - image with 4 dimension sub-cal123_ses-M12_STIR.nii.gz

[plbenveniste]

Reporting this issue here for documentation in the CanProCo repo.

Original issue (@NathanMolinier):

For some subjects in the session ses-M12 niftii images have 4 dimensions:
sub-cal123_ses-M12_STIR.nii.gz

Response from @valosekj:

It seems that this particular 4D image contains two consecutively acquired 3D volumes (notice the slight motion between the volumes):

Possible solutions:

• split the 4D image into two 3D images (e.g., using sct_image) and use only one of them
• average the 4D image across the fourth dimension (time). But this is probably suboptimal since the resulting image might be blurred due to the movement between the 3D volumes.

[plbenveniste]

Great catch, @NathanMolinier !
This explains why the nnUNet inference was crashing.

After running a script on the entire PSIR and STIR dataset to check if other images had 4 dimensions, I saw that this was the only problematic image.

Adding this file to the exclude_M0_M12_comparison.yml file for the current analysis before deciding on a solution for this problem.

[jcohenadad]

 split the 4D image into two 3D images (e.g., using sct_image) and use only one of them

I would go with this.