Dear FreeSurfer experts,

I have processed a group of subjects by the freesurfer v6 beta version  (with recent upgrade of mris_make_surfaces from 2016/11/21).

It seems to me that in contrast to v5.3 version (with the same nsigma_above 2, nsigma_below 3) the T2 pial refinement cuts out too much gray matter in regions where the gray matter has rather low intensity in T2 images. I have observed this behavior quite often in my subjects.

I have uploaded the example subject cut_gm_T2pial.tar.gz to the server. The problematic area has RAS location 3.9, -70, 0. Also see the screenshot.

My command line was:

mris_make_surfaces -nsigma_above 2 -nsigma_below 3 -aseg aseg.hires -filled filled.hires -wm wm.hires -mgz -sdir $my_subject_dir -orig white.deformed -nowhite -orig_white white.deformed -orig_pial pial.preT2.pass2 -T2 my_subject/mri/T2w_hires.norm -T1 T1w_hires.norm -output .postT2.pass2 my_subject rh

and for FS 5.3

mris_make_surfaces -nsigma_above 2 -nsigma_below 3 -aseg aseg.hires -filled filled.hires -wm wm.hires -mgz -sdir $my_subject_dir -orig white.deformed -nowhite -orig_white white.deformed -orig_pial pial.preT2.pass2 -T2 my_subject/mri/T2w_hires.norm -T1 T1w_hires.norm -output .postT2.pass2.FS_5.3 my_subject rh

Could you please comment on how to proceed?

It seems that in contrast to v5.3 it is necessary to increase nsigma_below value since the nsigma values seem to be more "sensitive" in v6beta version. I tried to increase nsigma_below (nsigma_below for T2 in recon-all is 5) but it seemed to me that it leads to "leak" of the pial surface out of the gray matter in other areas, mainly in regions adjacent to cerebellum. See the example leaking the pial surface to the cerebellum in left hemisphere with nsigma_below 5 in the second screenshot in attachment ( I have uploaded this subject as file cut_gm_T2pial2.tar.gz, his cut-out gray matter RAS location is -33.7, -18, 45, command line is similar as above, but for left hemisphere). Yellow surface is with nsigma_below 5, blue is nsigma_below 3. Btw, the -aseg option in mris_make_surfaces does not prevent to leak surfaces into the cerebellum?

Could also be the reason of these problems, that I used HCP scripts which do not do the T2 normalization with anatomy-informed mri_normalize (using aseg and white surface), but only global normalization and bias field correction?

Regards,

Antonin Skoch