MR Brain Images Segmentation Based On Affinity Propagation

Medical image segmentation plays an important role in medical image application, but meanwhile it is a classical difficulty in medical image processing and analysis,especially the segmentation such as Magnetic resonance image(MRI). Different brain tissues hybrid together and the boundary between them are blurry, moreover intensity inhomogeneities in MR images cause the brightness distribution of different physiological organizations overlap, which make the segmentation of MRI be more difficulty than other image.For intensity inhomogeneities are smooth and vary slowly, we employ local clustering method to eliminate the adverse impact of intensity inhomogeneities in MRI. This method is based on the theory of Shannon entropy, fuzzy clustering algorithm, affinity propagation, which avoids information loss that caused by the estimation and correction procedures for intensity inhomogeneities.In order to find the local region with uniform intensity inhomogeneities as much as possible, we use the Shannon entropy as a homogeneity measurement to optimize the local region. Because of the complexity of brain tissue structure, the local region optimized by Shannon entropy could not satisfy the assumption of local image modal completely: within a context, the three classes (white matter, gray matter cerebrospinal fluid) of tissues exist together and there are considerable pixels in each tissue class. According to this case, we use affinity propagation algorithm to segment the gray level of each local region independently, which greatly reduced the calculation cost. Via combining the gray level information and the corresponding pixels number, the diagonal elements of similarity matrix are given value assignment appropriately, meanwhile the clustering number is adjusted in iterative procedure, which effectively guide segmentation. If the cluster result is two classes, we use the cluster center of adjacent context with similar statistic distribution to classify the two classes into corresponding tissues.The efficacy of the proposed algorithm is demonstrated by extensive segmentation experiments using simulated and by comparison with other published algorithm.