I’m convinced it will never hurt to do so, and improve your results slightly anyway, so I would certainly advise it. However, don’t expect too much of it though: given a realistic CSF response, it will never introduce drastic changes to the FODs in white matter hyperintense lesions, since the CSF response at the b≠0 shell should only show very little signal. Or to put in in another way: it’ll only fully remove FODs of the size/kind you’d see in the actual CSF (in the ventricles) when doing simple single-shell single-tissue CSD.
I’ve got an ISMRM abstract accepted that relates very closely to this topic (the one I also mentioned in this and following posts). As mentioned over there, I’ll make the text available, e.g. on ResearchGate, ASAP after the proceedings are available for all attendees.
Another side note: you may still see your FODs “visually” (when viewing in mrview) more drastically changing in size when switching from single-shell single-tissue CSD to MSMT-CSD (e.g. with WM and CSF), but those changes will be visually prevalent over all (healthy) WM. This has nothing to do with the multi-tissue aspect, but rather with the hard non-negativity constraint of the MSMT-CSD implementation (whereas the original (SSST-)CSD implementation has a “soft” constraint/regularisation). The fact that this makes your FODs look smaller, is a bit misleading: they actually also become slightly wider, so it’s just the peak amplitude that becomes smaller. The lobes’ integrals will generally be very similar in magnitude.
Finally, as mentioned a few times in this thread above as well, in the future you’ll be able to use single-shell 3-tissue (SS3T) CSD as an option too, which can give you the full 3 tissue types from just single-shell + b=0 data. This will work better with higher b-values (e.g. 2500-3000 and up), and depends more crucially on very decent SNR for lower b-values. However, “in the future” is still a little bit of a wait for now… as I’m sadly occupied with way to many different things at the moment. 