Expert knowledge guided segmentation system for brain MRI

Abstract

This paper presents a fully automated segmentation method for extracting a variety of anatomical structures in magnetic resonance images (MRI). We have developed a segmentation system where maximum use is made of the available medical expertise, either in the form of implicit knowledge or of explicit information. A series of deformable templates (simplex meshes), initialized via the non-linear registration of a reference segmented MRI, are evolved in a rule-controlled framework and subject to various constraints, so as to maximize the achieved match over the target structures. Segmentation results on brain MRIs are discussed and compared against manual delineations.