I have a couple conceptual questions about the application of the ICA-based dual-regression technique for between-group comparisons. Apologies if this has already been covered, but I don't believe I have yet found a firm answer to the following questions.

1) Is it advisable to perform some kind of denoising prior to running dual regression? For instance, should one start by performing single-subject melodic and regression of noise components, or is this inadvisable since noise components will later be discarded at the group level?

2) How to deal with unequal sample sizes, particularly when sample size is low. For instance, I have groups of 6 (group 1) and 8 (group 2). If I include all subjects when running group ICA to determine my components of interest, and then regress these components on each individual subject, won't subsequent group comparisons using those components be biased? Should I perhaps identify group components on just 12 subjects (6 in each group), and then regress these components on all 14 subjects?

Thanks very much for any advice!

David