Wrong fieldmap correction in 23.2.0 using opposite PE images

[alexsayal]

What happened?

I am getting incorrect distortion correction results with the new release of fMRIPrep. Specifically, the new "Alignment between the anatomical reference of the fieldmap and the target EPI" step seems very off:

This generates a bold corrected image with some weird spots and warps:

In a different subject, it is a bit worse:

I am using two spin-echo images with opposite phase encoding directions (AP and PA). This error was not found for version 23.1.2.

Do you have any suggestions? Thank you!

What command did you use?

udocker run \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2:/data \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2/derivatives/fmriprep2320:/out \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2/derivatives/fmriprepwork:/work \
    -v /users3/uccibit/alexsayal/freesurfer_license.txt:/licensefile \
    fmriprep2320 \
    --participant_label $SUBJECT \
    --fs-license-file /licensefile \
    /data /out participant \
    -w /work \
    --stop-on-first-crash \
    --output-spaces MNI152NLin2009cAsym:res-2 \
    --mem=85G \
    --skip_bids_validation

What version of fMRIPrep are you running?

23.2.0

How are you running fMRIPrep?

Docker

Is your data BIDS valid?

Yes

Are you reusing any previously computed results?

No

Please copy and paste any relevant log output.

No response

Additional information / screenshots

No response

[effigies]

Can you share screenshots of the before/after images? GitHub doesn't like to show animated SVGs.

[alexsayal]

Here they are for the first subject:

[effigies]

Thanks. That's definitely a bad boldref/fmapref alignment, probably due to such different contrasts in your BOLD and your fieldmap inputs. What images are used as inputs to your fieldmap estimation for this BOLD series?

[alexsayal]

I am using two spin-echo EPI images, known as Spin-Echo Fieldmaps in human connectome project nomenclature. Their geometry and echo spacing matches the functional BOLD images. I can send an example if it helps.

[effigies]

A listing of the files for this subject would be fine. It sounds like something like:

sub-01/
  fmap/
    sub-01_dir-AP_epi.nii.gz  # single-band
    sub-01_dir-PA_epi.nii.gz  # single-band
  func/
    sub-01_task-sometask_bold.nii.gz  # dir-AP, multiband

If your fieldmaps are already well-aligned to your BOLD, you could try passing in a --derivatives directory containing an identity fmapreg file:

$PRECOMPUTED/
  sub-01/
    func/
      sub-01_task-sometask_from-boldref_to-auto00000_mode-image_xfm.txt

(You'll need to specify the correct fieldmap identifier. You should be able to find the one associated with your BOLD image in your output directory.)

You would then rerun with:

fmriprep $INPUT $OUTPUT participant --derivatives fmapreg=$PRECOMPUTED ...

[alexsayal]

The following command does not work - it assumes 'fmapreg' is part of the path. What am I missing?

udocker run \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2:/data \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2/derivatives/fmriprep2320:/out \
    -v /users3/uccibit/alexsayal/BIDS-BRAINPLAYBACK-TASK2/derivatives/fmriprepwork:/work \
    -v /users3/uccibit/alexsayal/freesurfer_license.txt:/licensefile \
    fmriprep2320 \
    --participant_label $SUBJECT \
    --fs-license-file /licensefile \
    /data /out participant \
    -w /work \
    --stop-on-first-crash \
    --output-spaces MNI152NLin2009cAsym:res-2 \
    --mem=85G \
    --skip_bids_validation \
    --derivatives fmapreg=/out

[alexsayal]

Also, you are right about the file listing. Now I have added the transformation matrices as such:

[effigies]

Ah, for some reason I thought that we used named derivatives, but it looks like it's just a list of paths.

You can use --derivatives /out, though I would personally not mix up my inputs and outputs like that. I would probably organize things like:

fmriprep/
    sourcedata/
        raw/
        freesurfer/
        fmapreg/

Then from within fmriprep:

fmriprep sourcedata/raw . participant --derivatives sourcedata/fmapreg ...

You're of course welcome to organize things however you like, but I get paranoid about telling a program that it can write out to one of the input directories.

[alexsayal]

Thank you, it is now working as expected. I still think the final image is better corrected and aligned to the anatomical with version 23.1.2, so I will keep using that version for now.
One possiblity to explore is to invert the colors of the SPE image before entering fmriPrep to see if it helps the registration to the functional EPI. Do you have any hints for this? This correction algorithm was optimized for images of the same type, right?