Dear Joelle,
Alright, I thought it was task fMRI as most of the paradigms discussed on the list are task-based. For resting state data it is going to depend on type of analysis for sure, e.g. with ICA one would probably not regress out WM or CSF signal while one might well want to do so for e.g. seed-based analyses. As stated, various preprocessing strategies and algorithms have been proposed, and while there is a recent review on motion correction by Power et al. (2015, http://dx.doi.org/10.1016/j.neuroimage.2014.10.044) I'm not aware of a consensus regarding other aspects / nuisance regression. E.g. global signal regression is disputed, some studies have reported only modest use of WM/CSF signal regression (e.g. a recent study by Shirer et al., 2015, http://dx.doi.org/10.1016/j.neuroimage.2015.05.015 ) while others have highlighted the advantage. Different findings might also depend on type of analysis and the particular implementation of the denoising (average signal vs. principal components across all WM voxels above a certain threshold vs. a WM at a particular coordiate vs. an eroded mask), especially as often, a certain strategy does not just account for WM/CSF signal but some additional parameters. Which doesn't answer your question I'm afraid, but there's probably no simple answer.
Best
Helmut
|