Quantitative Study of Brain Anatomy

We introduce a system that automatically segments and classifies features in 3-D images. The system’s accuracy is comparable to manual segmentation. It takes 12 minutes to segment and classify 144 brain structures in 256x256x124 voxel image, while similar work by human took 8 months. The process starts with an atlas, a hand segmented and classified MRI of a normal brain. Given a subject’s data, the atlas is warped in 3-D using a hierarchical deformable matching algorithm until it closely matches the subject, i.e. the atlas is customized for the subject. The customized atlas contains the segmentation and classification of the subject’s anatomical structures. The system has processed MRI of 105 subjects, and for 97 of them produced segmentation qualitatively comparable to manual segmentation. We performed quantitative evaluation of the precision for one structure. Of 18 subjects for which classification correctness was examined voxel by voxel, an error rate of less than 20% was achieved for 12 subjects, and less than 10% for 5 subjects. The efficiency, accuracy, and consistency of the system’s performance allow for detailed quantitative studies of brain structures. Initial results have been obtained for finding the normal range of variation in the size and symmetry properties of anatomical structures, and for detecting abnormalities.