Artificial intelligence–based assessment of breast arterial calcification on screening mammography and cardiovascular outcomes: a meta-analysis
M Malik, P G Batista, L A Lucena, M V Montenegro, R R A Santos Silva, G C Gaelzer, C Fischer Bacca, J GiorgiAbstract
Background
Cardiovascular disease remains the leading cause of mortality in women, yet existing risk stratification tools frequently underestimate cardiovascular risk, particularly in younger and ostensibly low-risk populations. Breast arterial calcification (BAC), an incidental finding on routine screening mammography, has emerged as a potential marker of cardiovascular risk. Recent advances in artificial intelligence (AI) enable automated, quantitative BAC assessment, but the prognostic significance of AI-derived BAC across cardiovascular outcomes has not been systematically synthesized.
Purpose
To evaluate the association between AI-based BAC detection on screening mammography and cardiovascular outcomes in women.
Methods
PubMed, Embase, and Cochrane databases were systematically searched from inception through December 2025 for observational studies assessing AI- or deep learning–based BAC quantification on full-field digital mammography. Eligible studies included women without prior major adverse cardiovascular events (MACE). The primary endpoint was MACE; secondary endpoints included all-cause mortality. Hazard ratios (HRs) were pooled using inverse-variance random-effects models, and event rates were synthesized using generalized linear mixed models. Heterogeneity was assessed using I² statistics, and risk of bias was evaluated using the QUIPS tool.
Results
Three retrospective cohort studies comprising 134,326 women were included. Mean or median age ranged from 56.8 to 69 years. AI-detected BAC was associated with a nonsignificant trend toward higher MACE risk (pooled HR 1.21, 95% CI 0.94–1.56; I² = 97.8%), with pooled MACE incidence of 13.53% (95% CI 7.67–22.75%). In contrast, AI-detected BAC was significantly associated with increased all-cause mortality (pooled HR 1.31, 95% CI 1.04–1.66; I² = 91.8%). Sensitivity analyses confirmed the robustness of these findings, although substantial heterogeneity persisted.
Conclusions
AI-enabled BAC assessment on screening mammography is associated with increased all-cause mortality and shows a consistent, albeit nonsignificant, association with MACE. These findings support the role of mammography-based AI BAC analysis as a scalable, opportunistic tool for cardiovascular risk stratification in women.Graphical Abstract