Research On Encephalic Tissue Recognition For MR Image Based On Support Vector Machine

Medical image recognition is the precondition of image visual in special tissues. It is also a method of extracted quantitative information. Image recognition has widely applied in the diseased tissue diagnosis, the localization and anatomical study, computer guidance operation, etc. The encephalic tissue recognition and tumor recognition for MR images have been studied in this paper.For MR image, it is not enough for encephalic tissue and tumor recognition if only depended on gray degree. So more features have to be extracted, the support vector machine depends heavily on selection of feature vectors and the spatial distribution. In this paper, in order to acquire high accuracy of recognition, the extracting feature method of combined textures and gray features are researched.Statistical learning theory is a kind of special research based on small sample case of machine learning theory. With the development of its theory, statistical learning theory has been gotten more and more extensive attention. Statistical learning theory is based on a set of relatively solid theoretical foundation. To solve the problem of learning finite sample, Statistical learning theory offered a unified framework. There are many existing methods. The theory is expected to solve many original difficult problem (such as neural network structure selection problem, the local minimum points problem, etc.), Meanwhile, a new general learning tool - support vector machine generated which is based on this theory, it has already showed a lot better results than existed machine learning methods. In Encephalic Tissue MR images, intracranial various tissues of the boundary are extremely complex and irregular, they are highly nonlinear. The traditional recognition algorithm got the challenges. Support vector machine has great advantages to solve the problem of the high-dimension, nonlinear and irregular classification. In this paper, support vector machine is used to classified different tissues due to its advantages: It can effectively solve the problem of small sample, the non-linear and the high-dimension. Its learning process can get global optimum. The model established from learning process has good generalization ability. However, selecting parameters for SVM is a complicated problem. It is blind to select parameters manually. Immune algorithm is introduced to search the best parameters for SVM to solve the problem. Immune Support Vector Machine (ISVM) is proposed to classify encephalic tissue to reach higher classification accuracy by combined SVM and immune algorithm,Immune Feature Weighted Support Vector Machine (IFWSVM) which was based on Immune Support Vector Machine was proposed to further improve the classification accuracy of MR image. Considered that each dimension should have different contribution to the final result, IFWSVM deals with this problem with weighting dimensions.In this paper, three algorithms of traditional support vector machine, immune support vector machine, immune feature weighted support vector machine were used to recognize the encephalic tissue, which include the five kinds of normal tissue: background, cerebrospinal fluid, bone dense, cerebral white matter, brain gray matter, one kind of abnormal tissue: tumor recognition. The experimental results show that the immune weighted feature support vector machine approach can reach higher classification accuracy than immune SVM and traditional SVM. This method can be widely applied in the field of classification for MR recognition.