In automatic subcortical segmentation, each voxel in the normalized brain volume is assigned one of about 40 labels, including:
- Cerebral White Matter, Cerebral Cortex, Lateral Ventricle, Inferior Lateral Ventricle, Cerebellum White Matter, Cerebellum Cortex, Thalamus, Caudate, Putamen, Pallidum, Hippocampus, Amygdala, Lesion, Accumbens area, Vessel, Third Ventricle, Fourth Ventricle, Brain Stem, Cerebrospinal Fluid
FreeSurfer now runs automated labeling of the brain volume and this is included in the stable v3.0 release, during the -autorecon2 stage. However, if you processed your anatomical data using previous versions and you wish to obtain the automated labels, you can just run the subcortical segmentation separately.
To obtain automatically segmented volumes for the first time (assuming the -autorecon1 stage has completed), run:
recon-all -subcortseg -subjid <subject name>
The automatic subcortical segmentation can take many (11+) hours to complete.
It is important to note that the stable v3.0 release of FreeSurfer by default uses the automatically segmented brain volume (ASEG - in mri/aseg.mgz) to segment the white matter volume (WM - mri/wm). In doing this the edits are different than the manual edits you may have already done in older versions of freesurfer (i.e., the ventricles are filled in completely) this creates a different folding pattern and uses different atlases. You should be sure that all of your subjects are edited in the same manner, using the same atlases (i.e., don't mix freesurfer versions). If you wish to process one subject further (i.e., to get final surfaces) and wish to maintain your edits you can use the -noedit_wm_with_aseg flag along with the -old-atlas flag. This will preserve your manual edits and utilize the old atlases. This is not ideal, as you should process all your subjects fully with the same version of FreeSurfer.
To view the segmentation, use this command:
tkmedit <subject name> norm.mgz -segmentation mri/aseg.mgz $FREESURFER_HOME/FreeSurferColorLUT.txt
Float your cursor over any voxel and the label assigned to it will be displayed in the TkMeditTools window.
If the voxels are incorrectly labeled (this is more likely to happen along the medial aspect of the brain), then you can re-label them yourself. Refer to the TkMeditGuide/TkMeditWorkingWithData/TkMeditSelectionsLabels page for detailed information on this procedure.
Automatic subcortical segmentation of a brain volume is based upon the existence of an atlas containing probablistic information on the location of structures. This is decribed here:
Whole Brain Segmentation: Automated Labeling of Neuroanatomical Structures in the Human Brain, Fischl et al., (2002). Neuron, 33:341-355.
The atlas included with the Freesurfer distribution is found in the 'average' directory, and is called 'RB_all_2008-03-26.gca'. It is possible to construct your own atlas. This is described next.
Constructing an Aseg Atlas
- Using tkmedit, label each volume in the brain. Repeat for all subjects to be included in the atlas.
Run mri_ca_train to create the atlas.
See also rebuild_gca_atlas.csh script in $FREESURFER_HOME/bin.
See also AtlasSubjects