Research On Segmentation Of MRI Human Brain Images

Magnetic resonance imaging(MRI)is an effective and noninvasive approach to observe human brain. Now MRI is widely applied in many fields, such as medicine, neuroscience, psychology and learning-science, et al, because of its own advantage. The main target of MRI brain image segmentation is to delineate profiles of neuroanatomical structures and also label each voxel as belonging to a specific structure.Traditional gray-level histogram is of 1D, and all spatial information is discarded after the histogram was built. The joint-histogram of gray-level and gradient contains spatial information of local neighbor, so it can utilize the image information more sufficiently. After sketching regions of certain gray-level and gradient, according to brain tissues(GM & WM), on the 2D joint-histogram, the region growing algorithm is then implemented to wipe off those tissues not belonging to brain introduced by the thresholding process mentioned above.The process of linear seeds diffusion is to plant a mass of seeds into regions of interest(ROI) in the image, then to obtain segmentation result just according to the seeds distribution which is the solution of the linear diffusion equation at last. This method is employed to segment brain tissues from MRI image directly, with implicit difference discretization and AOS algorithm applied.Conventional smoothing methods not only eliminate noise but also blur boundaries in the image. The substance of nonlinear diffusion filtering is: 1)region boundaries keep their positions; 2) intra-region smoothing occurs preferentially over inter-region smoothing at any scale. Here the Weickert's multi-dimensional CLMC model is solved by implicit difference discretization and AOS algorithm, and its stability is validated through experiments on 2D MRI image. Then this model is employed on the smoothing of a MRI image as a pre-processing method, and the WM is finally segmented by the region growing algorithm.