Problem with registration between fieldmaps and functional images (v23.2.1)

Summary of what happened:

Dear fmriprep experts,

after updating to version 23.2.1 we sometimes encouter problems with the registration of fieldmaps to functional images. This happened in some rare cases (roughly 3 out of 200 functional runs) in our projects. We have a Siemens MAGNETOM Prisma Scanner, and use two magnitude images and the phasediff image for sdc. We would be grateful for any help or ideas!


Command used (and if a helper script was used, a link to the helper script or the command generated):

singularity run \
    --bind ${INPUT}:/data \
    --bind ${WORK_DIR}:/work \
    --bind ${OUTPUT}:/out \
    --cleanenv \
    --clean-workdir \
    --skip-bids-validation \
    --participant-label $SUBJID \
    --md-only-boilerplate \
    --fs-license-file /data/LICENSE.txt \
    --nthreads $nthreads \
    --stop-on-first-crash \
    --cifti-output 91k \
    --use-aroma \
    -w /work \
    /data /out participant



Environment (Docker, Singularity / Apptainer, custom installation):


Screenshots / relevant information:

This is the alignment under version 23.2.1:

This is the alignment under version 20.2.1:

I think this may be the same as Noticeable difference in scanner-computed fieldmap results since LTS · Issue #413 · nipreps/sdcflows · GitHub. Would you be willing to share this subject as we work on this?

I have not had a great deal of time to work on this problem lately, but I hope to get back to it soon.

Hello there,

I am facing the same issue where the registration between my BOLD ref and fieldmap ref is quite off for a few subjects. I am working 7T data ¶ and used opposite polar (PA and AP) spin echo images for PEPOLAR distortion correction. This registration has been okay for most of the runs, but is quite off for some. When I checked the alignment between the respective xxx_desc-coreg_boldref.nii.gz and xxxfmapid-auto00000_desc-epi_fieldmap.nii.gz by loading them separately in fsleyes, they seemed fairly well aligned, and certainly unlike what I was seeing in the output html reports.

Would be quite valuable to have some insights into what might be going wrong, and what can be done to correct it. I am also using v23.2.1