Hi Paul,
yes, we know this happens - it's because of the way the manual training set was constructed, so that it has voxels labeled cortex there that shouldn't exist. The accumbens in general isn't all that accurate as it has no real intensity signatures at its boundaries, so we don't use it a lot. It's on our list of things to fix.
Bruce
On Wed, 24 Jan 2007, Paul Greenberg wrote:
Hello,
I'm having trouble getting an accurate subcortical segmentation (aseg.mgz, and aparc+aseg.mgz) in the Left Hemisphere Nucleus Accumbens area of one subject. Specifically, regions that appear to be the Left Nucleus Accumbens are being labeled as Left Cerebral Cortex. The two images (attached) illustrate the problem. Image 1 shows only pial and white matter surfaces on the "brainmask.mgz" file while image 2 shows the pial and white matter surfaces on "brainmask.mgz" along with the aparc+aseg.mgz segmentation. For this subject I added control points to fix other inaccuracies in other regions, and edited the wm.mgz to cover all regions I considered basal ganglia (the white matter surface still appears to invade regions of the basal ganglia). I reprocessed this data with recon-all -autorecon2-cp -autorecon3, but am still not satisfied with the results. What would you suggest as the best way to more accurately identify the accumbens regions in this case?
Thanks much!, Paul