This study evaluates the accuracy and feasibility of using artificial intelligence (AI) for the detection and assessment of breast arterial calcifications (BAC) on mammography in a large screening population across 15 sites in a prospective study.

Sequentially accrued 2D mammograms from 15,785 asymptomatic screening women during a 1-month period (3/23/2023-4/19/23) across 15 screening sites were analyzed using a deep learning AI algorithm specifically designed for BAC detection. Age of the women ranged from 20 to 97, with a median of 56. The study assessed overall prevalence of BAC, as well as distribution of BAC across four age groups: < 50, 50-59, 60-69, and = 70. The AI model was trained using an internal Real World Dataset of 2D mammograms to detect BAC based on expert annotation and provides an assessment score of 0-5 for BAC according to the total area of BAC and its density. Prior to the study, the accuracy of the AI algorithm is validated on a validation dataset of 2D mammograms from 8,898 women. There is no overlap between the training, validation and 15,785 women prospective study datasets.

The AI algorithm achieved an area under the ROC curve (AUC) of 0.938 (95% CI: 0.928 - 0.949) on the validation dataset, indicating high accuracy in BAC detection. In the prospective study with 15,785 women, the overall prevalence of BAC detected by the AI algorithm was 14.9% (95% CI: 14.5% - 15.3%), with a prevalence of 4.0% (95% CI: 3.6% - 4.4%) in women < 50, 8.8% (95% CI: 8,1% - 9.4%) in women 50-59, 19.8% (95% CI: 18.7% - 20.8%) in women 60-69, and 40.8% (95% CI: 39.3% - 42.3%) in women = 70 with a BAC score cutoff of 2 and above.

AI based BAC detection on mammography in a large screening population across 15 sites is feasible and accurate.The AI algorithm demonstrated high accuracy in BAC detection, with a prevalence and distribution of BAC increasing with age as expected in a screening population. Our results suggest that AI can standardize BAC detection at scale, potentially improving efficiency and reducing inter-observer variability. The age-specific prevalence of BAC provided by our study can inform clinical decision-making and risk assessment with an expectation of patient volumes.

Breast arterial calcium is gaining popularity as a proxy for global atherosclerotic disease. Accurate quantitative models for automated BAC assessment from mammograms can offer an adjunct screening tool for heart disease.