A tightly coupled region-shape framework for 3D medical image segmentation

Most hybrid 3D segmentation methods either heuristically couple the respective algorithm or combine a true 3D with a 2D algorithm due to computational considerations. In this paper we propose a new probabilistic framework for 3D image segmentation that combines tightly linked region- and shape-based constraints. Region-based label constraints are modeled by a 3D Markov random field, and are tightly coupled to shape-based constraints of a 3D deformable model. The full 3D nature of the combined model leads to a robust smooth surface segmentation that outperforms the single constraint, slice-based as well as the loosely coupled 3D methods