Soft computing approach to brain structure segmentation

This paper presents a soft computing approach to automatically segment subcortical brain structures from MR head scans. The technique combines spatial information from anatomy and tissue characteristics of the structures. The spatial locations of the structures are learned by an enhanced fuzzy-MLP network. The evidences of a particular voxel belonging to a particular structure, rendered by spatial and intensity information are combined by a fuzzy fusion operator. The results on automatically segmenting hippocampus, putamen, caudate, amygdale, and globus pallidus on an ensemble of head scans are shown.