I have never tried, and as I mentioned above, this is doable but I think labor intensive. A more direct alternative would be to use the same mixed model (ie a non-deformed template) for all subjects. You would loose in anatomical accuracy but that would make the group analyses straightforward.