Support for limited field of view

[chrisgorgo]

To improve temporal and spatial resolutions sometimes a limited field of view is used. In such case only preselected part of the brain is imaged. This can lead to challenges when coregistering to T1w volume. We can use the 7T TRT dataset to test any potential solution.

[oesteban]

Several people asked for this in the context of MRIQC as well :+1:

[oesteban]

Ref #899, #900, #901, #896, #895.

Make the appropriate amends to allow FMRIPREP to work correctly on these images:

• [ ] Revise the reference and mask calculation workflow
• [ ] Inpainting for co-registration?

[suxpert]

Are there any update on this topic? What is the best solution for limited FoV scans nowadays?

[milenamusial]

I'd be interested, too!

[effigies]

If your BOLD and T1w data are already in rough alignment (open as overlays in some viewer), then my recommendation would be --bold2anat-init header, which will skip the bulk registration and only perform the boundary-based registration refinement.

If they are not mostly aligned already, then you may want to use a tool like 3dslicer to manually align first.

[natalia-zaretskaya]

We are having cases where limited field-of-view data are well-aligned originally, or pre-aligned manually, but running --bold2anat-init header throws off the original alignment and the data end up being misaligned (coregistration fails miserably without this flag).

These are images of the occipital slab acquired with left-right phase-encoding, so there is not much of the cortex for bbregister to work with, and quite a bit of cerebellum. I thought that an old school cost function like mutual information could work better, and tried running mri_coreg from FreeSurfer (which I guess is what recent fmriprep versions are using?) with structural as target and functional as movable, adding --regheader flag. This seems to work well for the few subjects I tried.

        mri_coreg \
        --ref $refvol \
        --mov $nifti \
        --reg $nifti_dir_out/${filename}.lta --regheader

Another alternative that seems to help is additionally passing the flag --no-pass1 to bbregister:

    bbregister --s sub-${sid} \
    --mov $data_dir/sub-${sid}/ses-1/func/sub-${sid}_ses-1_task-CFE_run-${run}_bold.nii.gz \
    --reg tmp_nopass1.lta \
    --init-header --bold --no-pass1

I would find it useful if users could pass some custom options to mri_coreg and/or bbregister, similar to how the expert options file is implemented in FreeSurfer's recon-all. Is this something one could consider for the future?

We are happy to provide data for testing.