Advancing breast cancer risk assessment with digital breast tomosynthesis

Effective treatment schemes have reduced breast cancer mortality, but precise identification of women at increased risk of developing breast cancer, to personalize screening and preventive interventions, accordingly, is a major outstanding challenge. This urgent clinical need is most prominent among Black women, in view of their higher breast cancer mortality than White women. Studies have consistently shown the potential of artificial intelligence in the form of deep learning (DL) to elucidate novel mammographic signatures highly predictive of breast cancer. However, currently available DL models were developed in mostly White populations, and therefore, may generalize poorly in Black populations that would benefit from risk-based tailored screening and preventive strategies. Moreover, most related DL studies rely on digital mammography (DM) images, although digital breast tomosynthesis (DBT) has rapidly replaced DM in the US. These studies also paid limited attention to model interpretability, which is critical for clinical translation of DL models. We will leverage a unique multi-site resource of screening data from Black women (4507 cases; 90,701 controls; with 5-year follow-up) and state-of-the-art DL and medical imaging informatics tools, with the aim to enable accurate long-term (i.e., 5-year) breast cancer risk assessment for Black women. We will accomplish this through developments of DL imaging signatures of breast cancer risk with the new standard of breast cancer screening, DBT; thorough evaluations of their clinical utility in a multi-site setting; and deployment/dissemination activities to enable further evaluations and refinements based on feedback. We propose three aims. SA1 will develop a DBT-driven breast cancer risk score via DL and its combination with the clinical Black Women’s Health Study (BWHS) risk model into a hybrid breast cancer risk prediction tool. SA2 will perform independent clinical utility evaluations of our breast cancer risk prediction tool. SA3 will focus on translational innovation, by developing the deployment framework and disseminating our tools, knowledge, and resources. Our study will be the first of its kind on computational mammographic signatures of breast cancer risk among Black women. We anticipate that the successful completion of our aims will provide a novel tool for accurate long-term breast cancer risk assessment among Black women, extensive multi-site validation data and a complementary deployment/dissemination framework to enable further evaluations in research settings. We expect that the groundbreaking outcomes of this project will have a major impact on advancing personalized risk assessment for Black women, which, in turn, will lay the foundation for enhanced precision breast cancer screening and prevention strategies.