A population-based study of deep learning measures of breast arterial calcification for improving women's health using repurposed mammogram images - To achieve the goals articulated in the NIH Advancing Science for the Health of Women strategic plan and advance our understanding of the antecedents of chronic disease in women, innovative strategies that leverage longitudinal data in large diverse populations are required. Cardiovascular disease (CVD) is the single largest killer of women. Clinically used risk scores used to guide treatment and prevention strategies to delay or prevent CVD have variable performance across demographic characteristics. Tools that provide more precise estimates of individual risk in women could significantly reduce the risk of CVD. Numerous clinical and imaging biomarkers are routinely measured serially across the lifespan but are typically confined to screening, diagnosing, or monitoring single diseases. Breast arterial calcification (BAC) is a mammographic biomarker that is associated with CVD. This application focuses on detecting and quantifying BAC and studying the utility of BAC for predicting CVD in women. Screening rates of mammography are high throughout the nation in every demographic group and thus represent a unique opportunity to directly interact with women who might otherwise underutilize preventive health care services. In comparison with 2D mammogram, over 80% of the imaging sites now also acquire digital breast tomosynthesis (3D) images. Our overarching hypothesis is that automated quantification of BAC from 2D and 3D mammograms will allow opportunistic prognostication of CVD risk. Therefore, we propose to study 125,519 women from two geographically and socioeconomically diverse cohorts from the University of Mississippi Medical Center and the Upper Midwest region via the Rochester Epidemiology Project. Our access to currently available clinical and imaging data uniquely positions us to 1) extend and validate the AI algorithm to detect and quantify the extent of BAC on both 2D and 3D breast mammograms; 2) identify clinical, demographic, and socioeconomic factors associated with BAC prevalence and progression; and 3) assess the predictive value of BAC as an independent biomarker of cardiovascular risk in aging women. The infrastructure in place at each site to extract and process images will enable the assessment of BAC in over 500,000 index and serial mammograms. Finally, we will follow these cohorts for over a decade for the development of CVD risk factors and events to determine the impact of BAC prevalence and progression across the lifespan on subsequent disease. If our proposed studies show that BAC is a significant early predictor of CVD in women, BAC measures could become an important way to refer women to preventive CVD therapies at the time of mammography. As such, mammogram screenings could become a way to not only address risk of breast cancer but also become part of primary CVD prevention.
