A Digital Mammograhpy and Digital Breast Tomosynthesis Combined Method for Breast Cancer Classification

Digital mammography (DM) and digital breast tomosynthesis (DBT) are two effective radiological examinations for detecting breast cancer. DM is efficient in observing calcifications but limited in showing critical details of certain lesions due to glandular occlusion. DBT excels at visualizing subtle lesions, such as architectural distortions, that are often obscure in DM. However, DBT is hindered by imaging noise and information redundancy. In clinical practice, radiologists typically combine DM and DBT to provide a more accurate assessment of lesion malignancy. Most existing computer-aided approaches for breast cancer classification rely solely on either DM or DBT. Some recent studies attempt to combine DM and DBT for breast level malignancy analysis, but these methods are often coarse and lack interpretability. To address this issue, we propose a DM-guided DBT frame selection module to identify informative lesion frames in DBT and utilized these frames for DM and DBT combined malignancy analysis. Additionally, we implement a type-aware attention block to adjust the weights of DM and DBT features within lesion type, leading to more accurate pathology predictions. Experimental results demonstrated that our proposed method achieved superior performance with an AUC of 0.911 compared to utilizing only DM (0.845) or only DBT (0.822). We have also made the MammoLesion dataset publicly accessible to support further research. https://github.com/DM-DBT-Combined-Method-For-Breast-Cancer-Classification