By the way, I noticed something interesting - I have a subject with a stroke brain for whom I get really bad results with 3dSkullStrip. But I also ran 3dDWItoDT on the same subject, including the automask option, and it actually did a really good job stripping the brain. Maybe because the contrast for diffusion weighting is different than the T1 anatomicals that we usually use for our high-res anatomy.
The lesion area shows up really dark on T1, which is probably what is leading to large portions of brain being stripped out. The lesions are bright on T2 images however. So maybe a good approach for stripping stroke brains would be to acquire a T2, coregister / resample it to your high res T1 space, strip the T2, and apply the mask to the T1.
I tried running 3dSkullStrip on some T2flair and FSET2 images and it did a fine job. So I would just need to work out the scripts for the coregistration, deobliquing, regridding, etc. to use this approach, since my T1 anatomicals are much higher resolution and acquired with different slice orientation than my T2s (only acquired for clinical purposes, but they are turning out to be useful!) I might modify my scan acquisition strategy in the future to accomodate for this.
-Jed