Anatomical-driven segmentation of the 3/sup rd/ and 4/sup th/ ventricles in MR data

A method for automatic segmentation of the brain's 3/sup rd/ and 4/sup th/ ventricles in MRI (magnetic resonance imaging) datasets is introduced. The method exploits anatomical knowledge about these structures and uses gradient-based edge detection and volume-growing to complete the segmentation. Nearby anatomic landmarks, including the longitudinal fissure, cerebellum and callosum are also automatically extracted in our approach. The method has been tested on a variety of T1- and T2- weighted MR images.