Hi Woogul,
Either of those ways to make a mask is fine. Basically,
you just want a mask that you consider appropriate for
your group analysis (including ROI masks).
Your second question is an alternative to the first, isn't
it? Or do you mean that you have already masked each
subject's data in orig space?
If you have already masked the data, then the only really
fair way is to take the intersection of those masks in tlrc
space. Note that the union would not be good, since the
edge voxels would have data for only some of the subjects,
and the sample sizes would not be equal between groups.
For the same reason, a single subject's mask would not
be appropriate.
- rick