Quantitative Comparison of Coronary Artery Calcium and AI-Based Plaque Burden: Agreement, Discordance, and Clinical Implications

Background Coronary artery calcium (CAC) scoring quantifies calcified atherosclerosis but does not capture non-calcified plaque. Artificial intelligence (AI) – based quantitative coronary CT angiography (CCTA) enables the measurement of total plaque burden (TPB), reflecting both calcified and non-calcified components. The degree of agreement between CAC and TPB categories remains unclear. Methods We retrospectively analyzed 1,955 subjects who underwent both CAC scoring and quantitative CCTA. Subjects with CAC = 0 were excluded. CAC was categorized as 1–99, 100–299, and ≥ 300 Agatston units, while TPB was categorized as 0, 1–250, 251–750, and > 750 mm³. Agreement between CAC and TPB categories was assessed using Cohen’s weighted Kappa, and Spearman correlation evaluated the continuous relationship between CAC and TPB. Analyses were stratified by age, sex, diabetes, hypertension, and use of statins. Results Participants had a mean age of 66.5 ± 10.4 years, and 69% were male. CAC distribution was 1–99 (39%), 100–299 (20%), and ≥ 300 (41%); TPB distribution was 1–250 (67%), 251–750 (27%), and > 750 mm³ (12%). Weighted Kappa was 0.495 (95% CI 0.471–0.519), indicating moderate categorical agreement, while Spearman correlation was robust (ρ = 0.92). Agreement improved modestly when CAC cutoffs were adjusted to 1–99, 100–399, and ≥ 400 (κ = 0.535). Concordance was higher in males (κ = 0.53) than in females (κ = 0.37). Conclusions CAC and AI-derived total plaque burden demonstrate moderate categorical agreement but very strong continuous correlation, highlighting that quantitative plaque analysis captures complementary non-calcified atherosclerosis not reflected by CAC alone.