Dear experts,
I was rerunning recon-all on quite large group of subjects previously reconstructed and edited with v5.3. I was disappointed to see quite large amount of new surface errors in subjects after rerun with v6.0 which need many new manual time consuming interventions.
Most pronounced issue seems to me worse determination of wm.mgz, i.e. more regions of gray matter and "dura" is left in wm.mgz voxels. This is not primarily related to skullstrip, since I used brainmask.mgz from v5.3 for recon-all in v6.0 (to keep edits). The possible reason of worse wm.mgz is probably related by the filtering during normalization steps: I found that there is much more aggressive signal intensity filtering during normalization in v.6.0. See the screenshots with brain.mgz and white surface in one of the problematic regions. The white surface is OK with v5.3, whereas white surface is leaking out of the brain in v6.0 and there is much lower variance of signal in white matter regions in v6.0 in comparison to v5.3.
Could you please comment on?
Regards,
Antonin Skoch