Hi Ana,
You should apply a mask that says where the brain
is (or where data without major signal dropout is).
Such a mask does not have to be perfect, it merely
restricts the blur to a reasonable region.
Also, as is done with afni_proc.py, compute the blur
on a per-run basis, to keep detrending appropriate.
You might consider running afni_proc.py on one or
two subjects, both to compare the sequence of steps
and to verify the results.
- rick