If you do not have too many subjects and are concerned with the quality of the signal you get from
the NAC, you may consider making a different anatomical mask for each subject. From my
experience there is quite a lot of variability between subjects wrt NAC, and depending on how you
normalize the subjects' brains, you may find that a general mask does not overlap all that well with
the NAC of any of them...
(in my data the anatomical masks had not a single voxel in common for all 16 subjects, using the
standard SPM normalization).