Hi All, as some of you have figured out, we have a beta version of FS 8.0.0. Please give it a spin and let us know if you have any problems or concerns.

The release notes are here https://surfer.nmr.mgh.harvard.edu/fswiki/ReleaseNotes which I have copied below.
One bug that we have already found is that you must set environment variable FS_ALLOW_DEEP to 1, eg, export FS_ALLOW_DEEP=1 (for bash) or setenv FS_ALLOW_DEEP 1 (for csh/tcsh)

thanks
doug

Now uses SynthSeg, SynthStrip, and SynthMorph (deep learning algorithms). These make recon-all more robust and much faster. Technically, SynthMorph is not needed for recon-all to run, but the nonlinear registration can be used in multimodal analysis to map to the MNI152.
There is now DL automatic segmentation of regions that often cause errors in pial surface placement. These include the major sinuses (sagittal and transverse) as well as a spot between frontal and temporal lobes where the middle cerebral artery (MCA) and dura cause bright areas.
Several fixes in entorhinal cortex. First, the WM in EC is segmented using DL. The WM is often only 0.5mm thick and very hard to see; this often resulted in severe white surface areas in this location requiring manual editing to fix. The new surface now represents gyrus ambiens much better. Second, The white surface is extended to the front of amygdala. Third, the binary cortex of label now covers entorhinal cortex and parts of the temporal pole better allowing the pial surface to extend out to its true edge.
Some bugs where found in and around inferior frontal cortex causes the white surface to extend too far into basal ganglia. This has been fixed. This is above and beyond the issue reported in 7.3.2 below.
All the bug fixes described below are now turned on by default.
These changes will work with longitudinal with the caveat that the volumetric segmentation will not be fully longitudinal. The segmentation will be applied to the time point in the base space, but it will not (and cannot) be initialized with the base segmentation. Initial tests indicated that the change in power was minimal and mixed.
recon-all will now run in about 2h vs 8h on a single CPU. It will, however, required about 24GB of memory to run.
The use of SynthStrip should make recon-all much more robust as the old skull stripping method often included extra cerebral tissue.
With synthseg, we are now supplying a segmentation-based total intracrainial volume (sTIV) along with the registration-based estimated TIV (eTIV)
We have applied v8 to several studies (aging, AD crossectional, AD longitudinnal, schizophrenia). There were no major changes in the ability to detect effects, and there was no interaction between effect and version. But let us know how it performs on your studies.