Hi Douglas,
I am not sure why my previous message (posted Dec 22) did not get posted.
I have reported many problems with the skull stripping and resulting issues in freesurfer. To summarize:
1. For our structural scans, we have significant pial matter/dura matter remaining for > 50% of subjects or more (of a dataset of 350 scans). I have tested using freesurfer's gcut and other options, as well as AFNI skull strip, FSL with many options, and even ANTS.
2. We gave up on trying to use surfaces with this dataset, due to the sheer amount of manual intervention that would be needed to edit the brains. A colleague suggested that Freesurfer v 6 may work better than 5.3 , which we had been using. Do you think that would make sense?
3. The aseg almost always labels the remaining skull as gray matter, which then influences the icv and wbv measurements. 4. Recently we have decided to try the longitudinal pipeline. Another problem appears: if a cross-sectional scan does not have any pial matter (as in, it is one of the few that as successfully skull stripped - we call these the "clean" scans) but a template is made from multiple scans from this subject. If one of those other tp scans from this subject has any pial mater remaining, then the (a) the template will include all of the pial matter and (b) the longitudinal version of the "clean" scan now has extra pial matter that gets erroneously labeled as gray matter, or sometimes white matter.
I have been trying to solve this for over a year, and I also sent data to freesurfer over the ftp on December 1st using the directions on the freesurfer wiki, if you want to take a look at a single representative scan with this skull strip issue.
Thanks!
Michelle