Automatic compartmental decomposition for 3D MR images of human brain

A new method is developed to automatically decompose human brain MR images into cerebrum, cerebellum and brainstem. It utilizes the partial differential equations based shape bottlenecks algorithm to segment an interior brain tissue region into three compartmental seeds. Then the compartmental decomposition is obtained by reconstructing the compartments from the seeds according to the compartment boundary knowledge defined with partial volume information. This method was validated against manual segmentations of 35 T1-weighted MR images. It was demonstrated to be accurate and robust, and the mean Dice coefficients for cerebrum, cerebellum and brainstem were 0.99, 0.98 and 0.82, respectively.