CONTINUOUS VOXEL CLASSIFICATION BY STOCHASTIC RELAXATION - THEORY ANDAPPLICATION TO MR-IMAGING AND MR-ANGIOGRAPHY

In this paper we present a stochastic relaxation method for voxel clas sification in magnetic resonance (MR) images. This method is based on Bayesian decision theory. In this framework, the optimal classificatio n corresponds to the minimum of an objective function, which is here d efined as the expected number of misclassified voxels. The objective f unction encodes constraints according to two a priori models: the scen e model and the camera model. The scene model reflects a priori knowle dge of anatomy and morphology; the camera model relates observed MR-im age intensities to anatomical objects. Both models are described using the concept of Markov random fields (MRF). This allows continuity and local contextual constraints to be easily modelled via the associated Gibbs Potential Functions. The minimum of the objective function is a pproximated asymptotically by stochastically sampling the associated G ibbs posterior joint probability distribution. The method is applied t o brain tissue classification in MRI and blood vessel classification i n MR angiograms. Each application contains a novel aspect: in the form er, we introduce topological constraints on neighbouring tissues; in t he latter, we incorporate shape constraints on cylindrical structures.