Automatic Sub Classification of Benign Breast Tumor

Abstract

This paper describes about automatic classification of benign lesions. Our re-search work mostly aims at development of a data driven effective classification system to determine type of benign breast tumors. The existing system in medical field identifies the type of benign breast tumor based on histopathological report obtained after a painful surgical process. Our target is to develop a system that would work without histopathological attributes to avoid pains to the patient. Our focus was to eliminate the role of histopathological attributes in the detection of benign tumor type. So we tried to identify correlation of histopathological features with mammographic image features and patient history features in order to explore if histopathological features can be replaced by corresponding correlated features. With replaced attributes we gain training accuracy for J48 as 79.78% and with SVM 81.91%. We obtained testing accuracy for J48 and SVM as 100% and 90.90% respectively.