A Volumetric Analysis of Coronary Calcification on Non-Electrocardiogram-Gated Chest Computed Tomography Using Commercially Available Deep-Learning Artificial Intelligence

Objective: We examined the accuracy of coronary calcification volume (CV) measurements using deep-learning artificial intelligence (AI) for non-electrocardiogram (ECG)-gated chest computed tomography (CT) and compared it with the accuracy of the CV measured with a commercially available workstation (WS) and the Agatston score (AS). We showed the potential, limitations, and optimization of AI for evaluating coronary artery disorders. Materials and methods: Overall, 315 of 344 patients were analyzed. All patients underwent non-ECG-gated and ECG-gated non-enhanced CT during preoperative chest screening and/or chest pain assessment from March 7, 2021, to March 7, 2022. The accuracy of CV-AI was compared with that of CV-WS and the AS. Stratification grades based on CV-AI were compared with grades based on the AS. Cases of mismatched stratification were examined to determine the limitations of AI. The cut-off value with the best stratification of CV-AI was obtained. Results: The correlation coefficients between CV-AI and CV-WS and the AS were 0.964 (p < 0.01) and 0.960 (p < 0.01), respectively. Stratification of coronary risks showed significant consistency between methods (p < 0.01), and categories were matched in 81.0% of cases. When the AS was regarded as the “gold standard”, the accuracy, sensitivity, specificity, negative and positive predictive values, and Dice and Jaccard indices were 0.946, 0.921, 1, 0.856, 1, 0.959, and 0.921, respectively. AI rarely overlooked calcifications to underestimate coronary risks. The best cut-off values for categorization were 10-100-360 (default: 10-100-500). Conclusion: AI has sufficient potential to stratify the risk of coronary events on non-ECG-gated chest CT, particularly for non-cardiac patients. However, the results of AI analyses should not be blindly accepted.