Peer Reviewed Journal via three different mandatory reviewing processes, since 2006, and, from September 2020, a fourth mandatory peer-editing has been added.
We present a fully automatic mixture model-based tissue classification of multispectral (T1- and T2-weighted) magnetic resonance (MR) brain images. Unlike the conventional hard classification with a unique label for each voxel, our method models a mixture to estimate the partial volumes (PV) of multiple tissue types within a voxel. A new Markov random field (MRF) model is proposed to reflect the spatial information of tissue mixtures. A mixture classification algorithm is performed by the maximum a posterior (MAP) criterion, where the expectation maximization (EM) algorithm is utilized to estimate model parameters. The algorithm interleaves segmentation with parameter estimation and improves classification in an iterative manner. The presented method is evaluated by clinical MR image datasets for quantification of brain volumes and multiple sclerosis (MS).