Objective This study evaluated (1) the reproducibility of 5 visual scoring systems for breast arterial calcifications (BACs) on mammography and their correlation with coronary artery calcifications (CACs), (2) their diagnostic accuracies, and (3) clinical applicability. Methods In this retrospective study, 90 women (median aged 57 years, interquartile range IQR 15) who underwent both mammography and cardiac CT were included. Breast arterial calcification was scored using 5 visual systems: (1) dichotomous (present/absent), (2) 4-level vessel count, (3) 4-level severity, (4) combined 4-level (severity + vessel count), and (5) an advanced 3-level scale (incorporating vessel count, length, and density). Diagnostic performance was evaluated via sensitivity, specificity, and area under the curve (AUC). Interobserver agreement was assessed with Cohen’s kappa; correlation with CAC (Agatston score) used Spearman’s rho. Results Breast arterial calcification was present in 17.8% (16/90) and showed a trend toward correlation with CAC (P = .05). The 4-level vessel count scale showed the strongest correlation with CAC (rho = 0.373; P <.001). The highest AUC (0.604) was achieved by the 4-level severity scale (sensitivity 44%, specificity 77%). Interobserver agreement was highest for the combined 4-level scale (κ = 0.835), followed by the 4-level vessel count (κ = 0.833), severity-only (κ = 0.819), 3-level vessel count (κ = 0.833), and the dichotomous scale (κ = 0.740). The dichotomous scale required the least time (mean 19 s/patient); the 3-level scale required the most (mean 51 s). The vessel count scale balanced accuracy, reproducibility, and efficiency (mean 34 s). Conclusion Grading BACs is feasible, with consistently high interobserver agreement across scoring systems. The 4-level vessel count scale demonstrated a favorable balance of accuracy, reproducibility, and practicality, suggesting it may warrant further evaluation as a potential tool in mammography-based cardiovascular risk assessment.
Saenger et al. (Wed,) studied this question.