Purple rain: T1 segmentation fails but no errors

Hi there,
We’ve been using fmriprep with docker for a while but now we are trying to get it running on our HPC with Singularity. Although fmriprep (version 20.1.1) finishes successfully (or so it claims), it is somehow unable to get the T1w segmentation correct (and that produces a cascade of failures downstream). On the bright side the images look pretty with that purple background…

See here the bad segmentation:

Also here the full html report:

And this is a my last (minimal) call; the ones that include the functional images also fail:

    singularity --debug run \
    -B /path-to-home/:/home/fmriprep \
    -B /path-to-BIDS/:/data \
    -B /path-to-output/:/output \
    -B /path-to-license/:/lic \
    -B /path-to-temp/:/working_dir \
    -B /path-to-templates/:/templateflow \
    --home /home/fmriprep \
    fmriprep-20.1.1.simg \
    /data /output participant \
    --notrack \
    --output-spaces anat \
    --anat-only \
    --fs-no-reconall \
    --fs-license-file /lic/license.txt \
    --participant-label 101 \
    --work-dir /working_dir

Any idea of what could be the cause?

Hi Javier,

Brain extraction went definitely awry in this case. As a result, segmentation was run on the background (and that’s why you get the funny contours).

Could you share this dataset with us? Otherwise, would you be able to run antsBrainExtraction.sh on this dataset?

Hi there,
Sure. Find it here.

Hello again,

I am getting the same results when running Singularity locally (so it doesn’t seem to be a problem with our HPC configuration). Somehow it is working without problems on Docker on the same data set and with the same fmriprep version.

Any inputs?

Found the problem!
In case anyone running into the same issue, this seems to happen when the templates from templateflow are somehow corrupted. I deleted the old templates, pulled them again from templateflow and re-upload them to our cluster.

No everything is running fine again,