Topology preserving brain tissue segmentation using graph cuts

Xinyang Liu, Pierre-Louis Bazin, Aaron Carass, Jerry L. Prince

Abstract: In segmentation of magnetic resonance brain images, it is important to maintain topology of the segmented structures. In this work, we present a framework to segment multiple objects in a brain image while preserving the topology of each object as given in an initial topological template. The framework combines the advantages of digital topology and several existing techniques in graph cuts segmentation. The proposed technique can handle any given topology and enforces object-level relationships with little constraint over the geometry. We apply our algorithm to brain tissue segmentation and demonstrate its accuracy and computational efficiency.