Hi all,
A quick question about ICA-AROMA using fmriprep.
My goal is to preprocess some task-based fMRI data so that, ICA denoising is done and also nuisance regression is included. So, I was thinking to run fmriprep using the flag --use-aroma and then include in the GLM design matrix the estimated (I know it’s pre-aroma but its fine) nuisance regressors.
Based on “Box 3 Best practice guidelines” in doi.org/10.1038/s41593-020-00726-z :
While consensus on optimal preprocessing pipelines is lacking (see guideline #5), it is clear that nuisance regression of head motion, white matter and cerebrospinal fluid regressors, combined with low-pass filtering, is insufficient for denoising79,103. Instead, spatial ICA-based cleanup methods should be used where possible (and/or volume censoring where not).
First of all, I want to know briefly how exactly are the AROMA-cleaned data (~desc-smoothAROMAnonaggr_bold.nii.gz) built? How are noise components estimated and how they are excluded to finally output ~desc-smoothAROMAnonaggr_bold.nii.gz files ?
Next, assuming I get the ~desc-smoothAROMAnonaggr_bold.nii.gz files, which of the following scenarios would make more sense for the GLM design matrix construction?
a) trans_x, trans_y, trans_z, rot_x, rot_y, rot_z + DCT-basis regressors + the first 6 a_comp_cor_XX and t_comp_cor_XX
or
b) trans_x, trans_y, trans_z, rot_x, rot_y, rot_z + DCT-basis regressors + csf & white_matter global signals
c) any better alternative?