Automated parcellation of the brain surface generated from magnetic resonance images

We have developed a fast and reliable pipeline to automatically parcellate the cortical surface into sub-regions. The pipeline can be used to study brain changes associated with psychiatric and neurological disorders. First, a genus zero cortical surface for one hemisphere is generated from the magnetic resonance images at the parametric boundary of the white matter and the gray matter. Second, a hemisphere-specific surface atlas is registered to the cortical surface using geometry features mapped in the spherical domain. The deformation field is used to warp statistic labels from the atlas to the subject surface. The Dice index of the labeled surface area is used to evaluate the similarity between the automated labels with the manual labels on the subject. The average Dice across 24 regions on 14 testing subjects is 0.86. Alternative evaluations have also chosen to show the accuracy and flexibility of the present method. The point-wise accuracy of 14 testing subjects is above 86% in average. The experiment shows that the present method is highly consistent with FreeSurfer (>99% of the surface area), using the same set of labels.