AI-based opportunistic coronary artery calcium assessment on non-gated chest CT: a systematic review and meta-analysis

Abstract Coronary artery calcium (CAC) on routine non-gated chest CT can be quantified by artificial intelligence (AI), but performance at treatment-relevant thresholds and effects on preventive care are uncertain. We systematically reviewed 110 reports of AI-based CAC assessment on non-gated chest CT; 59 contributed to meta-analyses. Against paired ECG-gated CT, pooled sensitivity and specificity were 87.1% and 92.3% for CAC > 0 and 83.1% and 96.3% for CAC ≥ 100, with specificity above 97% at ≥ 300 and ≥ 400. Category disagreement was mainly underestimation: 30.7% of gated CAC 1–99 was classified as 0 and 40.3% of CAC 100–299 below 100. Adjusted hazard ratios for major adverse cardiovascular events rose from 1.60 (CAC 1–100) to 3.39 (> 400) versus CAC 0. In two randomised trials, AI-enabled notification increased six-month statin prescribing (51.2% vs 6.9% and 44.0% vs 9.8%). Substantial AI-detected CAC should prompt review of preventive treatment, whereas low AI-derived categories can underestimate disease burden.