Improvement in Ultrasound Breast Tumor Images Classification by A Support Vector Machine with Outliers Detection

• Main Authors: Su, Yu-Cheng, 蘇育成
• Other Authors: Wu, Wen-Jie
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/59507762821199926008

碩士 === 長庚大學 === 資訊管理研究所 === 94 === According to the statistics from the Department of Health, breast cancer is still on the top-5 list of the cause of death. Hence, regular health check and set up of a computer-aided diagnosis system is of great urgency. Owing to the shape of a benign tumor is quite different from a malignant tumor and is relatively stable that comparing with conventional features like texture, so it is not affected with different type of scanners. In this thesis, we classify a tumor according to its shape feature. The result shows that the classification result is good and the accuracy is 89.05%. Furthermore, we use a support vector machine (SVM) as the classifier. A traditional SVM is sensitive to the outlier in the samples. It affects the location of the decision hyper-plane and decreases the accuracy. Hence, we propose an effective method to solve this problem and will improve the performance of the SVM. The experiment shows that the proposed method successfully reduces the influence of the outliers and promotes the classification accuracy of a SVM form 89.05% to 91.43%.