Multi View Mammogram Analysis with Segmented Breast Masses and Calcifications for BI-RADS Classification

Breast cancer is considered as the most prevalent cancer among women worldwide. Computer Aided Diagnosis (CAD) systems have the ability to obtain promising results in early detection of breast cancer using different imaging techniques. This study focuses on developing a framework to classify segmented mammograms with breast masses and calcifications into the Breast Imaging-Reporting and Data System (BI-RADS) categories using multi view mammogram analysis. The proposed method consists of a dual path fine tuned EfficientNetB3 architecture for multi view feature extraction connected to an artificial neural network as the classification head. The suggested approach was evaluated using the INBreast dataset and it achieved an accuracy of 0.88, and a precision of 0.85 which is competitive with the existing methods which only performs the classification to BI-RADS based on a single type of abnormality. According to the results obtained, it can be observed that the proposed multi view approach to classify breast masses and calcifications into BI-RADS categories has an improved accuracy compared to the baseline single view CNN model showing its potential to aid in the early detection of breast cancer accurately. Also, the proposed method can be considered more clinically valid based on the fact that it takes the presence of different types of abnormalities instead of solely considering breast masses or calcifications in mammograms for the classification.