Validation of Artificial Intelligence-based Non-gated Chest CT for Coronary Artery Calcium Scoring Across Multiple CT Scanners.

Authors

Abstract

This study aims to validate the performance of artificial intelligence (AI)-based non-gated chest computed tomography (CT) for quantification and risk stratification of coronary artery calcification across multiple CT scanners. Patients who underwent both coronary computed tomography angiography (CCTA) and non-gated plain chest CT within one month were retrospectively enrolled. With the electrocardiogram (ECG)-gated coronary artery calcification measurement results as the gold standard, the coronary artery calcification volume, equivalent mass, and calcification score derived from non-gated chest CT with different scanners and parameters were detected. Spearman correlation coefficient (r), Bland-Altman analysis and intraclass correlation coefficient (ICC) were used to evaluate the correlation and consistency. ICC and Bland-Altman served as primary metrics for quantitative consistency; Spearman correlation was treated as supplementary association index. Kappa analysis and re-classification statistics were adopted to assess the consistency of coronary artery calcification (CAC) risk stratification, and a P-value less than 0.05 was considered statistically significant. For the overall population, the intraclass correlation coefficient (ICC) values of calcification volume, equivalent mass, and calcification score between AI-assisted non-ECG-gated chest CT and ECG-gat-d examination were 0.975 (95% CI 0.973-0.976, p < 0.001), 0.886 (95% CI 0.874-0.897, p < 0.001) and 0.972 (95% CI 0.910-0.927, p < 0.001). The corresponding Spearman correlation coefficients (r) were 0.974, 0.891, and 0.972, respectively (all p < 0.001). Bland-Altman analysis revealed proportional bias, with absolute measurement discrepancies increasing alongside rising CAC magnitude. The 95% limits of agreement (LOA) for each quantitative index were reported. The correlation coefficients (r) of different scanners were all above 0.8, and those of calcification volume and calcification score were both higher than 0.95. Risk-stratification consistency assessed by Kappa was good to excellent across scanner groups, and re-classification rates were documented. AI-based non-gated chest CT exhibits favorable consistency for coronary artery calcification quantification and risk stratification across multiple CT platforms, despite observed proportional measurement bias at higher calcification burden.

Tags

Topics