Automatic Multimodal Brain-Tumor Segmentation

Brain-tumor segmentation method is an important clinical requirement for the brain-tumor diagnosis and the radiotherapy planning. But the number of clusters is very difficult to define for high diversity in the appearance of tumor tissue among the different patients and the ambiguous boundaries about the lesions. In our study, the nonparametric mixture of Dirichlet process (MDP) model is used to segment the tumor images automatically, which can be performed without initialization of the clustering number. Furthermore, the anisotropic diffusion and Markov random field (MRF) smooth constraint are both proposed in our study. Our segmentation results for the multimodal MR glioma image sequences showed the properties, such as accuracy and computing speed about our algorithm demonstrates very impressive.