AI-Based Low-Dose Chest CT Assessment of Coronary Artery Calcification Score: An Analysis of Accuracy-Relating Factors.
Authors
Affiliations (6)
Affiliations (6)
- School of Medical Imaging, Fujian Medical University, Fuzhou, 350122, China.
- Department of Radiology, Fujian Medical University Union Hospital, Fuzhou, 350001, China.
- Department of Pathology, Shengli Clinical Medical College of Fujian Medical University, Fujian Provincial Hospital, Fuzhou University Affiliated Provincial Hospital, Fuzhou, 350001, China.
- CT Imaging Research Center, GE HealthCare, Changsha, 410000, China.
- Department of Radiology, Fujian Medical University Union Hospital, Fuzhou, 350001, China. [email protected].
- Department of Radiology, Fujian Medical University Union Hospital, Fuzhou, 350001, China. [email protected].
Abstract
This study evaluates the accuracy of AI-based coronary artery calcium scoring on nongated low-dose chest CT and identifies factors affecting calcium risk classification, using ECG-gated noncontrast cardiac CT as the reference standard. This prospective study included 312 patients with chest pain who underwent ECG-gated noncontrast cardiac CT and additional low-dose chest CT (LDCT). LDCT images were reconstructed using pulmonary and standard kernels. AI software automatically quantified Agatston, mass, volume, and total coronary artery calcium scores (CACSs). Patients were stratified by Agatston-based risk category, heart rate (≤ 65 vs. > 65 bpm), and BMI (< 25 vs. ≥ 25 kg/m<sup>2</sup>). Agreement, systematic bias, and risk classification performance were assessed using Bland-Altman analysis, ICC analysis, Friedman tests with Bonferroni correction for multiple comparisons, linear regression, and confusion matrices. Pulmonary kernel LDCT produced Agatston scores comparable to ECG-gated CT, with no significant differences across heart rate, BMI, or calcium severity subgroups. Under this kernel, measurement bias showed weak but significant correlations with heart rate and BMI (r = 0.145 and r = 0.131, respectively; both P < 0.05). Overall, nongated LDCT-derived CACS correlated strongly with reference CACS (r > 0.94). The low heart rate/low-BMI subgroup showed the best risk stratification performance, with 80% accuracy and good agreement (κ = 0.72). AI-assisted CACS assessment on nongated low-dose chest CT may serve as an adjunctive tool for opportunistic coronary artery calcification screening and preliminary cardiovascular risk stratification. Its accuracy may be influenced by reconstruction kernel and calcification burden, whereas heart rate and BMI may not be significant influencing factors.Trial Registration: Chinese Clinical Trial Registry (ChiCTR2500100073,2 April 2025, the study was retrospectively registered).