Automated Quantification of Mammographic Density in Craniocaudal Oblique-View Radiographs

Breast density plays a pivotal role in assessing breast cancer risk, influencing both the likelihood of developing the disease and the potential for radiologists to miss small lesions in mammograms. Therefore, reliable and automated methods for analysing breast tissue density are essential for radiologists. In this study, we propose a U-Net neural network trained on 170 craniocaudal oblique-viewed mammograms to identify breast tissue. The trained model is then employed to segment breast tissue and estimate surface density using the pixel count method. Our results demonstrate high accuracy for both breast area segmentation (98%) and breast tissue segmentation (97%), with IoU scores of 0.93 and 0.83, respectively. Moreover, we compare our technique to existing state-of-the-art methods and confirm its superior performance.