Re: [Freesurfer] Subcortical Segmentation Failures with Freesurfer v. 5.10

Hi Joshua

if you can upload one dataset where the new version is poor and the old one is good I'll look into it.

Thanks for helping to track this down Bruce

On Tue, 17 Apr 2012, Joshua Lee wrote:

Hi All,

I thought I'd share my recent experiences with the newest Freesurfer v. 5.10 and subcortical segmentation.

I ran a large sample of brains using the default stream of Freesurfer v. 5.10. The resulting hippocampal segmentations were extremely poor, yet on the same sample of brains Freesurfer v. 4.xx did quite well.

Using tkregister, we discovered that the hippocampus is not being registered well enough. Nick Schmansky one wrote in another post on this list-serve that one difference between versions is that "talairach.xfm is needed to compute nu.mgz, [and] talairach stage needs to precede the nu_correct stage" He suggested using -use-mritotal in the recon-all command to overcome tal-check failures.

I decided to try using -use-mritotal with this sample of my data, despite tal-check not failing in the default stream. Resulting segmentations were now both realistic and accurate, and near identical to those produced by Freesurfer version 4.xx.

Have others experienced poor subcortical segmentation with current version of Freesurfer? I should note, that I am having this problem on another sample of MPRages of very good quality. I am concerned that future versions of Freesurfer will limit use of the old stream to situations in which tal-check fails.

Thanks you all.  Sincerely,

Joshua Lee