MRI Tissue Classification Using High-Resolution Bayesian Hidden Markov Normal Mixture Models

Magnetic resonance imaging (MRI) is used to identify the major tissues within a subject's brain. Classification is usually based on a single image providing one measurement for each volume element, or voxel, in a discretization of the brain. A simple model views each voxel as homogeneous, belonging entirely to one of the three major tissue types: gray matter, white matter, or cerebrospinal fluid. The measurements are normally distributed, with means and variances depending on the tissue types of their voxels. Because nearby voxels tend to be of the same tissue type, a Markov random field model can be used to capture the spatial similarity of voxels. A more realistic model takes into account the fact that some voxels are not homogeneous and contain tissues of more than one type. Our approach to this problem is to construct a higher-resolution image in which each voxel is divided into subvoxels and subvoxels are in turn assumed to be homogeneous and to follow a Markov random field model. In the present work we used a Bayesian hierarchical model to perform MRI tissue classification. Conditional independence was exploited to improve the speed of sampling. The subvoxel approach provides more accurate tissue classification and also allows more effective estimation of the proportion of major tissue types present in each voxel for both simulated and real datasets.