A probabilistic tissue classification algorithm is described for robust MR brain image segmentation in the presence of partial volume averaging. Our algorithm estimates the fractions of all tissue classes present in every voxel of an image. In this work, we discretize the fractional content of tissues in partial volume voxels, to obtain a finite number of mixtures. Every mixture has a label assigned to it, and the algorithm searches for the labeling that maximizes the posterior probability of the labeled image. A prior is defined to favor spatially continuous regions while taking into an account different tissue mixtures. We show that this extension of an existing partial volume clustering algorithm, [8], improves the quality of segmentation, without increasing the complexity of the procedure. The final result is the estimated fractional amount of each tissue type present within a voxel in addition to the label assigned to the voxel.
展开▼
