The GM labels of the IIT Human Brain Atlas (v.3) were constructed in a probabilistic approach. First, individual subjects were segmented with FreeSurfer. The labels from all subjects were then transformed to the space of the atlas. Finally, the spatially-transformed labels from all subjects were used to define the final labels based on the highest probability of occurrence of a label in a voxel. A probability map for each label is also provided (see the document describing the contents of the atlas in the "Docs" link).

Version 3.1 of the IIT Human Brain Atlas posted on 7/6/14 contains new probabilistic GM atlases for both Desikan and Destrieux labels. Both GM atlases were built using the same approach as for the GM atlas in v.3 (i.e. FreeSurfer of individual subjects, transformation of labels to the space of the atlas, calculation of the probability of occurrence of a label in a voxel). More details on the available resources for both probabilistic GM atlases (with Desikan or Destrieux labels) can be found in the "Docs".