A Study On MRI Brain Tumor Segmentation Approach

Magnetic Resonance Imaging (MRI) is an important medical diagnostic tool, because of its high quality in image display, which now has been widely applied to detect pathological lesions and diseases in various tissues and organs, especially in tumor detection. The segmentation of tumor is important and of great significance in clinical and scientific research. Although many segmentation algorithms have been proposed, most have limitations. The dissertation focuses on the Fuzzy C Means (FCM) algorithm, Region Growing method, and a segmentation approach of brain tumor which combines the fuzzy affinity technology with the community structure detection technique based on the node similarity is proposed, and tries to improve the performance of them.Firstly, the dissertation briefly describes the characteristics and principles of many algorithms in medical image segmentation, also their advantages and disadvantages in MRI tumor segmentation. As the same time, the dissertation briefly describes the principle of MR imaging.Secondly, as the representative algorithm, FCM method is discussed emphatically in the dissertation. We analyze the parameter selection in FCM, and achieve the accurate tumor segmentation.Thirdly, to overcome the difficulty of manual threshold selection and sensitivity to noise in region growing method, an adaptive region growing method based on the gradients and variances along and inside of the boundary curve is proposed. Firstly, we use the anisotropic diffusion filter to preserve the edge information. Then the new model is given, which chooses the mean variance inside of the boundary curve and the reciprocal of the mean gradient along the curve as the research subject. The objective function of the model is to add two elements about gradient and variance mentioned. The minimum of the sum is the optimum result which corresponding to the desirable threshold. In region growing processing step, the threshold is increased gradually and the set of the coarse contour is obtained. Finally, through optimizing the model, the optimal segmentation result can be acquired from the set of contours. In clinical MRI image segmentation, our method can produce very satisfactory results. In the last, the method based on the fuzzy affinity depends on the initial conditions. An image is viewed as a network, and the detection of the community structure in networks is beneficial to understand the network structure and to analyze the network properties. The community structure is discovered based on node similarity, which is fast and efficient. Then a new segmentation approach of brain tumor which combines the fuzzy affinity technology with the community structure detection technique based on the node similarity is proposed in the dissertation.