Dear Michael,
yes, your impression is correct. AROMA is tailored for head motion, but you get some 'free denoising on the side' as some non-motion components get flagged e.g., on the frequency content feature. The phenomenon is mentioned in the papers, but we did not assess the extent to which this happens.
Depends on your data really. e.g., if there is some slice artefact, or clear breathing related components (e.g., in multiband data) I would still go ahead and remove those...
Some metrics to use would be in the ICA-AROMA evaluation paper. Yet, I always wonder why people are so anxious to show the efficiency of ICA-based noise removal, whereas they do not seem to care about this when adding a 'gazillion' of noise-related covariates to their model (given that the ICA-based noise removal operates in the exact same way...).
Cheers,
Maarten
