This is a fairly complicated topic really. Regarding the specific questions of atrophy measures, there are a few possibilities:
1. CSF probability. Using probabilistic segmentation from a program like 3dSeg or similar, you can generate probabilities of different tissue classes. Gray matter probability is another option here.
2. Deformation to a normal template with Jacobian to get a compression/expansion type metric.
[
afni.nimh.nih.gov]
I have more thoughts on this, but it's a bit much for a message board post.