Medical Image Segmentation based on a 3D-MRF

To 3D medical image, the 2D Markov random field (MRF) does not include z-direction ´s information. In this paper, we propose a 3D-MRF image model based on 2D MRF by extending 2D planar to 3D space, define and describe the 3D neighbor, clique and potential function. We segment medical image using the 3D-MRF and the steps are as follows: 1.Initial images are segmented by using k-means clustering, to reduce the computational burden by using a special data structure: the k-d tree. 2. The parameters are estimated by using the maximum a posteriori (MAP) for the 3D-MRF model. 3. Computing optimal Solution is done using the expectation-maximization (EM) algorithm and the iterated conditional models (ICM) algorithm. Experiments show the 3D-MRF includes more neighboring information and the results of segmentation are more stable and practical.