I apologize in advance if this question has been asked already in some way shape or form, as it seems pretty elementary, but I figured this is the best resource of experts I could consult!
I would like to use FMRIPREP on minimally preprocessed, publicly available data. One basic question I have is whether running motion correction again on already motion-corrected data will…do bad things, I guess. Other questions and ruminations are below.
ABCD says that the minimally preprocessed T1s have been corrected for gradient nonlinearity distortions and intensity inhomogeneity, resampled to 1mm isotropic and aligned to an in-house atlas, however, when they ran Freesurfer they bypassed the initial intensity scaling and N3 intensity inhomogeneity correction due to the in-house processes run beforehand. Functional data has been HMC-ed with 3dvolreg, corrected for B0 distortion and gradient nonlinearities, and “resampled with cubic interpolation into alignment with each other, correcting between-scan motion, using a scan in the middle of the session as the reference.”
Given that they used 3dvolreg and FMRIPREP uses MCFLIRT, I worry about doing motion correction again (and not having it be to a reference in middle of all scan sessions). Should I be? Also, will running INU again cause issues? I believe I can skip distortion correction in the functional stream, but I don’t see an option for the structural intensity normalization. Is there a way to turn off these modules of FMRIPREP, and just run from skull-stripping and alignment between T1 and functional onwards? What would experts recommend doing from this starting point?
Basically, I’d still like to recon these folks as I’d like to do surface-based analyses, and ABCD surfaces are not being released yet. Also, will want to regress out CSF, etc, so having those confound regressors will be helpful, and the downstream analysis pipeline I’d like to use works with FMRIPREP output, so I was looking for a way to plug these subjects into that easily.