Enhanced Spatial Priors for Segmentation of Magnetic Resonance Imagery

A Bayesian, model-based method for segmentation of Magnetic Resonance images is proposed. A discrete vector valued Markov Random Field model is used as a regularizing prior in a Bayesian classification algorithm to minimize the effect of salt-and-pepper noise common in clinical scans. The continuous Mean Field solution to the MRP is recovered using an Expectation-Maximization algorithm, and is a probabilistic segmentation of the image. A separate model is used to encode the relative geometry of structures, and as a spatially varying prior in the Bayesian classifier. Preliminary results are presented for the segmentation of white matter, gray matter, fluid, and fat in Gradient Echo MR images of the brain.