Hi,
I've mailed this once before and tried to solve a persistent problem. Reviewing the results of recon-all I noticed that the derived pial surface chopped away an unacceptable amount of gm voxels. First of all I tried to manually edit some wm voxels, however with no success. Then I tried another solution by combining the FSL-FAST wm segmentation with wm.mgz because I noticed that FSL_FAST produced better results. Although the results did improve, still the pial surface was incorrect. The last option I tried was another set of parameters for nu_correct. Because it concerns 3T scans I used --proto-iters1000 and --distance 50 (see paper of Zheng et al, NeuroImage 48, 73-83 (2009)) and previous e-mails on this subject ("nu-correct parameters", 27 April 2010). Also with these settings the pial surface chopped away too many voxels. I'm desperate, the manuals provide no options on how to fix such problems. The pial surface can be corrected by only REMOVING voxels that are included erroneously but there is no way to extent the pial surface by ADDING voxels.
Ed