Implementation of BI-RADS Classification and Priority Prediction for Mammogram Pre-screening based on Multi-decision Framework

Mammography is the general way for breast cancer screening worldwide. However, the amount of Negative breast cases is always more than that of Positive ones. It results in radiologists spend much time on the Negative mammograms in the earlier period of screening. This paper introduces a method to assess the risk of breast lesions by two-category BI-RADS. Meanwhile, this method finds out the Positive cases before the Negative ones. First, the detection module finds breast lesions and generates confidence scores to quantify the severity of malignant lesions. Subsequently, the analytics module performs a two-category BI-RADS classifier and then predicts the priority of the mammogram pre-screening based on the multi-decision framework. The experiment results demonstrate that the proposed method achieves higher accuracy of 76% than the compared approaches in two-category BI-RADS classification, and advances the mammogram pre-screening by the 15% number of breast cases at least.