I have a clinical imaging dataset that I am trying to run through the fMRIprep workflow but the process is not completing. The input data contains 3 subjects.
In the clinical imaging workflow, gadolinium is delivered prior to the MRI scan session to save time. This means the T1w sequences are gadolinium-enhanced. It seems the extra contrast in the T1w sequences is causing fMRIprep to hang.
I am running fMRIprep v1.5.4 as a singularity container on a remote HPC with the following parameters:
--nthreads 8 --n_cpus 8 --mem_mb 32000 -anat-only
I have tried adjusting the job resources but it still hangs after 72hrs. It appears to be stuck on recon-all.
I’m wondering what else I could try? Has anyone tried fMRIprep on gadolinium-enhanced T1w images?
I preprocessed similar images last week. For me, 23.2.0 finishes successfully although I specifically skipped the recon-all step since I do not need it.
It seems that creating a brain mask for these images is more difficult than regular t1w images as I had te rerun 1 out of 7 subjects that had a lot of non-brain tissue inside the mask, which made normalization fail completely. For the others, the dura is regularly included inside the mask.
I could imagine that freesurfer has issues with these things. But I haven’t explicitly tested that.
If the images look something like the ones here in Figure 1, then I would run a separate bias correction or unifizing step (in AFNI, that’s 3dUnifize or 3dLocalUnifize).
I have had great success using synthstrip to brain extract data regardless of modalitiy, including Gd enhanced T1 scans. I would suggest using synthstrip one your images, and subsequently running fmriprep with the --skull-strip-t1w skipargument.