Supported by the National Natural Science Foundation of China under Grant No.60736008 (国家自然科学基金)
This paper presents an effective automatic method of brain extraction through twice segmentations respectively. The noise of the image is eliminated through anisotropic diffusion filtering based on Catt model while the details of the image are kept. The over-segmentation problem of the watershed algorithm is solved based on the merging of gray-scale similarity regions and the brain tissue is initially segmented. The edges of different organizations of every brain image are fuzzy and the MRI data is vulnerable to be affected by noise, therefore the non-brain region may be mistaken for brain region in the process of merging. In order to solve these problems, the Level Set method is adopted. The outline of the watershed is taken as the initial curves of level set to realize the automatic segmentation of brain tissue. The feasibility and practicality of this algorithm is proved by the results in the experiments.