Hi Anita,
We suggest not masking at all until after single subject analysis.
It helps you notice artifacts and it helps to avoid computing group
results where only some of the subjects have data.
Whether individual subject analysis is done in native space does
not matter that way (we suggest doing it in standard space). You
can review my 'WARP TO TLRC NOTE' note from "afni_proc.py -help"
for comments on that subject. :)
Getting higher % change values at the edge of the brain is normal
because of partial volume effects and motion. But I would rather
see what is happening there (and try to understand it :) than mask
it out and not even know. Voxels that have this effect should not
end up as significant with respect to regressors of interest though
(which is part of why motion parameters are included in regression).
- rick
