Using BI-RADs Breast Lesion Features-Based Classification for Breast Detection in Ultrasound Images

The death rate among women increases day by day and cases of it increases every year in world. The breast radiologists are used various imaging technologies such as mammogram and ultrasound for early detection of breast cancer. For detection of breast lesion type in ultrasound image, various features such as shape, orientation, margin, echo pattern, texture and shadowing are used in real time scenarios. Many existed algorithms were used large number of features and most features are numeric values. It is difficult to interpret numeric values of each feature by radiologist in real time scenarios. In this paper, we have given information regarding Breast Imaging Reporting and Data System (BI-RADs) features of breast lesion in ultrasound image which are used for decision about type of lesion in real time scenarios. The BI-RADs features contain lesion shape, lesion orientation, lesion Echo pattern, lesion texture, lesion margin and lesion shadowing. Also, the paper gives performance of various machine learning based classifiers using these real time breast features for classification of breast lesion. The performance of classifiers shows that the accuracy achieving up to 85% using K-nearest neighbor classifier for breast lesion classification and indicates that effectiveness of various classifiers for implementation of real time application in early detection of breast cancer.