Applied Research, Support Vector Machines For The Classification Of Breast Masses

Breast cancer is one of the most common malignancies in women, and its incidence rate is the first place in the women malignant tumor. Breast cancer is increased year by year at the rate of 3% and showed with the tendency of a continuously decreasing of pathogenesis age. Therefore, diagnosis and prevention of breast cancer is given great attention to in medical field, and its serious harm to human health has paid a high attention on World Health Organization (WHO). If the cancer is early detected, the early diagnosis is beneficial to the choice of effective treatment method. But there are some difficult points in research on the breast tumor, such as the of the breast tumor size is indefinite, the contrast ratio is small and the breast tumor is surrounded by the background with the similar characteristic.In this thesis, based on the current research on diagnosis technique of the breast tumor, the application of support vector machine (SVM) to breast tumor recognition is deeply researched. In order to decrease noises, and not to increase fuzzy of breast tumor margin, the median filter with the edge-preserving is selected. Because there are some fuzzy edge in the breast picture, using the method of fuzzy contrast ratio enhancement the image is enhanced to meet the demand of doctor vision with satisfactory effect. In the segmentation of breast image masses, the breast image is segmented using fuzzy set, and the regions of interest are obtained. Furthermore, based on different texture characteristics showed by different breast masses, nine texture characteristic parameters, which be used as basis of mass classification, are extracted using gray level co-occurrence matrix. Finally, SVM is studied, including theoretical foundation of SVM, selection of kernel function and selection of key parameters, and the SVM is applied to classification of breast mass and the classifier of the breast mass on the basis of SVM is implemented.Finally, the classification of breast mass using designed classifier is experimentally studied to distinguish benign masses from malignant mass. Training and identifying SVM by selecting Gaussian kernel function and polynomial kernel function and using cross-validation method, the optimal parameter model is obtained and the significant experimental results are achieved, which lay a good foundation for further studies.