Siemens XA30: heudiconv fails to handle partial volumes

[mvdoc]

Whenever a run is manually halted, Siemens XA30 generates a valid enhanced DICOM with fewer slices. If that happens, dcm2niix splits the series into a nifti with the full volumes and another nifti with the partial volume (see rordenlab/dcm2niix#742). Currently, heudiconv doesn't handle this case, and crashes because it tries to overwrite the intact volume with the partial volume.

Possible solutions:

1. Since dcm2niix will flag the partial volumes as derived, a solution would be to allow calling dcm2niix with the flag -i y, which ignores all derived series. But this is not really a great solution, since there are cases where we may want to convert also derived series.
2. dcm2niix adds a derived string under the bidsguess key in the output json. We could check if that string exists, but I guess it would work only when heudiconv is run with the flag --bids
3. The partial volumes are valid nifti files with fewer slices than prescribed. If multiple files are generated from the same series, we could maybe add a check on the number of slices to detect the partial volumes.

[bpinsard]

I see 3 options to fix that issue:

• a heuristic specific dicom_filter that checks dicoms tags are reported in the issue above, needs to be used for all XA data
• heuristics to allow specifying per-series the parameters for dcm2niix call: in that case -i y .
• a more global fix that would remove the partial volumes after dcm2niix conversion in convert.save_converted_files, partial volumes would be detected if the same series produced 2 files, and only one with "RawImage": False metadata. see [FIX] remove XA partial volumes, generated by dcm2niix as derived data #817