Anatomical CompCor enhancement

[rciric]

• Allow anatomical CompCor to operate separately in WM and CSF compartments (w_comp_cor and c_comp_cor according to BEP012 and RC1). Whereas the original approach described by Behzadi and colleagues (and currently used in fmriprep) uses a combined WM+CSF noise ROI for anatomical CompCor, aCompCor as deployed more recently by Muschelli and colleagues uses separate nuisance masks for WM and CSF. This latter approach was also used in recent benchmarking studies (1, 2).
• Relatedly, some images may not have sufficient resolution for discrimination of separate WM and CSF masks. (For instance, much of the T2w rat data we’ve looked at has poorly discriminable CSF boundaries.) Thus, it would be helpful if the pipeline supported the option of using only the WM compartment when generating regressors.
• To facilitate selection of an appropriate number of components for confound regression, save the eigenvalues/singular values for each component identified by CompCor approaches. Percentage of variance explained can either be saved directly in an additional metadata field (preferable) or computed from the singular values. The draft of BEP012 contained provisions for doing this; the most logical place for this would probably be in ~desc-confounds_regressors.json or ~decomposition.json.

I would be happy to help with any of these implementations that you think are worthwhile.

[oesteban]

They are worthwhile indeed. Please go ahead :)

[oesteban]

@rciric is this done? Please close if so or describe what's missing otherwise