Application Of Support Vector Machine In Medical Image Segmentation

Medical image segmentation is an important application in the field of image segmentation, and it is also a classical difficult problem for researchers. Thousands of methods have been put forward to segment medical image. Some use classical methods and others use new mothods.Vapnik and his collaborators proposed a useful algorithm: support vector machines, which can implement the structural risk minimization principle in statistical learning theory. This novel algorithm handles the classification problems successfully. Since then more attentions have been paid to statistical learning theory and support vector machines. The attractive research inclouds the improvement or modification of support vector machines by optimization techniques, and the design of the novel non-linear machine learning algorithms based on statistical learning theory and some ideas in support vector machines, etc.Support vector machine (SVM) is to correctly classify samples into two parallel planes in input or feature space by optimal planes(lines) ,and the margin between the two classes is made to be the largest. The standard SVM requires solving quadratic program that needs considerably longer computational time. An algorithm is introduced to solve the problem in this thesis, which is successfully applied to the classification of medical image data.Segmentation of brain tissues is very important in medical image analysis. Support Vector Machines (SVM) is considered a good candidate because of its good generalization performance, especially for dataset with small number of samples in high dimensional feature space. This thesis investigates the segmentation of magnetic resonance brain tissues image based on SVM.The main work in the dissertation can be summarized as the following:(1) Have realized the application of SVM method in medical image segmentation. The experiment result based on simulation data and real MRI dataproved the validity of the method.(2) A SVM segmentation algorithm based on fuzzy training sets is proposed, and the experimental result has proved the efficiency of this method. Also, the future working direction was discussed.