fMRIprep skull-strip and registration error

1

Hi all!

I have a set of T1/rs-fMRI scans; the T1s have already been processed with freesurfer recon-all.

When running a subject, pointing fMRI prep to the pre-existing fs directory, the cross modality registration appeared erroneous. I suspect due to a faulty brain mask. Below is the command and the resulting T1/MNI registered volumes.

Initial command:

usr/pubsw/packages/fmriprep/21.0.1/fmriprep-21.0.1.simg
–skip_bids_validation
–participant_label “$SUBJ”
–output-spaces MNI152NLin2009cAsym fsaverage:den-10k
–cifti-output 91k
-w $OUT_DIR/fMRI_prep/fmriprep_tmp/
–fs-license-file $DATA_DIR/code/license.txt
–fs-subjects-dir $OUT_DIR/fs-dir/
$DATA_DIR/ $OUT_DIR/fMRI_prep/ participant -v

This completed with out errors, but resulted in a T1 brain mask that incorporated skull and mid face and an inaccurate T1/MNI registration.

I re-ran using a pre-skull stripped T1 (from freesurfer), and skipping the skull-strip step, as follows:

usr/pubsw/packages/fmriprep/21.0.1/fmriprep-21.0.1.simg
–skip_bids_validation
–participant_label “$SUBJ”
–output-spaces MNI152NLin2009cAsym fsaverage:den-10k
–cifti-output 91k
-w $OUT_DIR/fMRI_prep/fmriprep_tmp/
–fs-license-file $DATA_DIR/code/license.txt
–fs-subjects-dir $OUT_DIR/fs-dir/
–skull-strip-t1w skip
$DATA_DIR/ $OUT_DIR/fMRI_prep/ participant -v

Again the brain mask extended well beyond the brian and again the registration to MNI was inaccurate, as below:
Brain mask:

sub-1039_ses-1_dseg
sub-1039_ses-1_dseg1128×564 182 KB

I can supply the registration gifs if needed.

Any advice on how to either supply a brain mask or skip this step entirely?

Thank you in advance!
Sam Snider
MGH/BWH
Harvard Medical School

2

Hi Sam,

Which version of recon-all did you use? fMRIPrep is best designed to use version 6 outputs by default (though plans to move to 7.X in the future).

Do the T1s look like they have good tissue contrast? Is this occurance subject-specific or prevalent across all subjects?

Just to clarify, does this mean you put the FS-output T1 in the raw data folder? If so, I’d recommend against that.

If it isn’t too much trouble, as a test, can you try not reusing the previously computed recon-all and have fMRIPrep rerun it? When doing so, I’d recommend using a new work directory too.

usr/pubsw/packages/fmriprep/21.0.1/fmriprep-21.0.1.simg
–skip_bids_validation
–participant_label “$SUBJ”
–output-spaces MNI152NLin2009cAsym fsaverage:den-10k
–cifti-output 91k
-w $OUT_DIR/fMRI_prep/fmriprep_tmp_TEST/
–fs-license-file $DATA_DIR/code/license.txt
$DATA_DIR/ $OUT_DIR/fMRI_prep/ participant -v

Best,
Steven

3

Thank you Steven.

recon-all was completed with freesurfer version 7

Yes the T1 was good quality. Freesurfer had no trouble masking the brain.

Yes, I did this just a test, after converting to nifti format and reorienting.

The reason why I prefer to do it this way, is some of the brain scans have significant lesions. I have a custom pipeline in freesurfer that is more robust to the presence of lesions (uses samseg), and so would prefer to run it outside fMRI prep.

So I already know that recon-all will fail on a good portion of the subjects.

Do you have any other ideas? Thanks again.

4

Have you read this about providing fmriprep with lesion masks? Processing pipeline details — fmriprep version documentation

5

I will take a look. May be challenging to scale that approach.

If the cross-modality (T1/BOLD) registration looks accurate, does it matter if the BOLD_space-MNI volumes are misregistered, seeing as I will be using the .gii output?

Thanks!
Sam

6

If you are worried about the surface files, perhaps you should instead compare the fsaverage subject output surface to the fsaverage template surface. Since resampling/spatial normalization happens in a single step, perhaps looking at this will make it clear where fMRIPrep is going wrong. That is, if the surface registration looks fine, then T1/BOLD registration is probably okay. If both the MNI and fsavergae resamplings are poor, then that may implicate a poor T1/BOLD registration that is being propogated.