I am new to using fMRIprep and I have some questions when performing motion correction.
Is there a suggested motion correction cuttoff for quality control when using fMRIprep (and ICA AROMA specifically) (e.g. 1-5 mm of motion)? At what stage in the preprocessing pipeline should this be identified exactly? Would that be for example after the preprocessing is done or before fMRIprep is run? I have data that might have some movement artifacts.
Is a multiband sequence compatible with ICA AROMA?
I am rather new to preprocessing and to fMRIprep hence the above questions.
Don’t know about second point, but for point 1) no there’s not a standard motion cutoff. It depends on a variety of factors. Are you working with children or clinical populations? If so you may want a more lenient outlier threshold. Do you have a lot of subjects (e.g. huge public data base) and are willing to throw some out? Use a stricter threshold. You can run a power analysis to see how many subjects are needed to achieve a good effect size and use a threshold that gives you that sample size (assuming the threshold is not too low as to ignore motion).
The nice thing is that none of these considerations require you to rerun fmriprep. Motion outlier censoring and subject removal happen during denoising and QC, respectively, which is not done by fmriprep (and are usually the next two steps).
Thank you very much for your reply. This is really helpful. I was mainly asking about the ICA AROMA option as I was under the impression that it removes the artefacts when it is run (instead of keeping them as regressors to use in a GLM afterwards). Would that still be the case for the ICA AROMA?
ICA AROMA does return the “non-aggressively” denoised version, but there is an fmriprep argument to return the AROMA components as part of the confounds.tsv file to do more aggressive denoising. If you do AROMA, it is up to you if you want to remove motion outliers, but you can read more about denoising strategies here (Ciric et al., 2017)