Brain Tissue Classification

For brain MR images, our goal is to segment the image into $K = 4$ regions corresponding to the (a) white matter, (b) gray matter, (c) cerebrospinal fluid, and (d) all other tissue types. This section starts by giving a high-level version of the proposed iterative classification algorithm along with an initialization strategy. It gives a few ways of incorporating a priori information in the probabilistic atlases into the proposed method. It describes the details of an efficient strategy for choosing the Parzen-window sample $\mathcal{A}_t$, explains why the method performs reasonably well without explicit inhomogeneity correction, and describes a optimal data-driven choice of important internal parameters.