Hi JW,
Sorry we did not get back to you on this.
It is preferable to make the masks based on the functional data. Alternatively, you could start with the brain mask datasets output by FreeSurfer and interect, fractionally intersect or take the union of them using 3dmask_tool.
What we like to do is create masks from the functional data intersected with those of the anatomical images, just to keep them tight.
Note that masking can be applied after 3dMVM. It is nice to scrutinize group results for oddities or artifacts, and not just see masked results from the beginning. Masking could be applied as you correct for multiple comparisons.
Do the ASL volumes lend themselves to reasonable masking (e.g. with 3dAutomask)?
- rick