Add option to revert to old behaviour of grouping scans across different acquisitions?

[michellewang]

Summary

Hi, I have a dataset where some of the participants have multi-shell data split across multiple scans (one scan per b-value), and we used the acq entity to label them. For example the raw BIDS data would look something like this:

1. sub-001_ses-1_acq-B0_dir-AP_run-01_dwi.nii.gz
2. sub-001_ses-1_acq-B0_dir-AP_run-02_dwi.nii.gz
3. sub-001_ses-1_acq-B0_dir-AP_run-03_dwi.nii.gz
4. sub-001_ses-1_acq-B0_dir-AP_run-04_dwi.nii.gz
5. sub-001_ses-1_acq-B0_dir-PA_run-01_dwi.nii.gz
6. sub-001_ses-1_acq-B1000_dir-PA_run-01_dwi.nii.gz
7. sub-001_ses-1_acq-B2000_dir-PA_run-01_dwi.nii.gz
8. sub-001_ses-1_acq-B700_dir-PA_run-01_dwi.nii.gz

So there are four B0 scans in the AP direction, then 4 in the PA direction with b-values 0, 1000, 2000 and 700. All of them have accompanying .bval/.bvec files too.

PR #862 changed QSIPrep's behaviour to account for the acq tag during the distortion group definition. Unfortunately, this results in undesirable behaviour in 1.0.0rc1 in my case, where the scans are not being merged as expected. With 0.23.0 I was getting a single desc-preproc_dwi.nii.gz file, while with 1.0.0rc1 I am getting a desc-preproc_dwi.nii.gz file for each of the B1000, B2000, and B700 scans, and I think there was was an error for the B0 scans because only the JSON was produced (might be a related issue).

I understand the reasoning for grouping by acq, but would it be possible to add a flag that allows the user to revert to the old behaviour of ignoring the acq entity when defining the scan groups?

Happy to create a PR for this if you could point me to the parts of the code to change.

Additional details

Next steps

[tsalo]

@mattcieslak and I have been talking about it and we came to the conclusion that QSIPrep should use the MultipartID metadata field to define these kinds of groups.

The basic idea is that, if multiple DWI scans share a MultipartID value, then they will be combined. I don't think the ID applies across phase encoding directions, so in your case you would probably have two IDs: apgroup and pagroup (or something similar).

[michellewang]

Ok, I think that makes sense, and modifying fields in JSON sidecars is easier than renaming files in BIDS, so that would be doable for my dataset too.

Since MultipartID is optional, what would the default behaviour be? Treat every scan as a separate group?

[tsalo]

Yeah I think that should be the default behavior. I'm not 100% sure what QSIPrep will do with multiple scans that aren't delimited with acq and dir entities though. E.g., sub-01_run-01_dwi.nii.gz and sub-01_run-02_dwi.nii.gz.

[mattcieslak]

qsiprep relies on the phase encoding direction and total readout time in the sidecars to determine if the distortion is the same in a group of scans