The Study Of Basal Ganglia Segmentation Algorithms Based On Brain MRI

Medical Image Segmentation has been the classic problem in the field of image processing. The brain has a very complex structure, and if we want to engage in the effective follow-up steps of treatment, segmentation in brain tissue is especially critical.Magnetic Resonance Imaging (MRI) is highly similar to soft-tissue resolution and imaging anatomy atlas-like display. The basal ganglia's main function is to exercise self-control and also involved in memory, emotion and reward. Basal ganglia's lesions can cause a variety of exercise and cognitive dysfunction, including Parkinson's disease and Huntington's, etc.. But as yet, we have not seen an effective partition for the brain anatomy with many sub-organizations, such as basal ganglia area.First of all, this paper briefly describes the characteristics and principles of many current common methods of medical image segmentation's algorithms. Then I put some representative algorithms to be used in the basal ganglia area, and discussed the results' non-adaptive of the area.Secondly, in this paper, a novel segmentation method, which is a combination of regional growth and morphology, is proposed for the brain tissue of MRI image. The algorithm uses region-growing as the central algorithm, combined with a priori knowledge and various reasonable morphological algorithms amendment as the follow-up processing steps. They complement each other, and deal with segmentation barriers from various angles to achieve the eventual realization of the effective partition for the basal ganglia in a complex background. Compared to the traditional region growing algorithm, the proposed algorithm uses a variety of methods to select seed points to reduce the segmentation results' dependence of the seed point. It also uses many morphological approaches to reduce the sensitivity on the local-maximum, and redundant partition under complex background.Finally, according to the characteristics of Gaussian kernel Fuzzy Clustering Algorithm and its good interface, I propose a novel segmentation method with the combination of co-occurrence matrix texture. This algorithm takes into account the texture characteristics of the target region and initial contour model of the expertise. In order to eliminate the disadvantage of the discontinuous classification, I also use the morphology as a follow-up step to extract a meaningful segmentation results. Because the algorithm adopts the result of the C-means clustering as the initial representative point, so compared with the traditional fuzzy clustering methods, it improves the accuracy and also accelerates the speed of segmentation because of the focus on the target area.