Tissue segmentation of multi-channel brain images with inhomogeneity correction

We propose a novel method to segment multi-channel magnetic resonance brain images into tissue classes taking into consideration the bias fields created by in-homogeneities of the scanners. The joint probability of tissue intensities in the multi-channel image is modeled using a multivariate Gaussian function; the prior models of tissue classes are presumed to be Markov random fields. An iterative algorithm is proposed to find the maximum a posteriori estimation of segmentation; suboptimally. Experiments on simultaneously acquired proton-density and T₂-weighted images are demonstrated.