Hi all,
I have a clarifying question about how fmriprep treats T1w images that had already been skull stripped. Specifically I am trying to understand if when using the flag –skull-strip-t1w skip , the input images still undergo bias correction with N4BiasFieldCorrection or any other kind of preprocessing before running freesurfer recon-all. My reasoning is that they shouldn’t, but I would love to confirm if that’s really the case, and if not is there a different way to control this behavior such that the input T1w will enter freesurfer without any further processing.
The Methods section in the html report describes bias correction as well as skull stripping with ANTS, even though I had skipped this step (it looks like fmriprep did skip this step but the report does not reflect that). I am running fmriprep version 23.0.2 through singularity on an HPC cluster.
Here’s the command I’m using:
singularity run --cleanenv /fmriprep-23.0.2.simg /input /derivatives participant --participant-label sub-xxx -w /work_fmriprep --output-spaces anat fsnative fsaverage MNI152NLin2009cAsym --dummy-scans 4 --ignore sbref --skull-strip-t1w skip --output-layout legacy --bids-filter-file /filter_T1w_skullstrip.json
Indeed you should be able to input pre-stripped images with the —skull-strip-t1 skip flag. The report generator may just have to be fixed to update that. Does the output t1 look okay and not overly eroded when you use this flag?
Thanks @Steven! yes it looks fine! the output brain has exactly the same mask as the input, so no additional masking/extraction was done. However the image contrast looks slightly different than the input, so I was wondering if it’s doing bias correction again or anything that could affect the accuracy of the freesurfer segmentations.
Good to hear that the image is not eroded / skull stripping is not repeated. I think the flag only disables skull stripping, so other steps such as bias correction should still be done. I think the main use case is for skull stripping raw images and using those as fMRIPrep inputs. Does that help?
Ideally I would want to have a flag to skip any preprocessing carried out on the anatomical image, so from your answer I understand this does not exist at the moment?
Not explicitly, but what you might be able to do in theory, is make a fmriprep-like output anatomical folder by running --anat-only, replace the final T1 with one you want to use, and then run fMRIPrep again using the anatomical fast-track option (--anat-derivatives <PATH>). It is not a very elegant solution though.