Accuracy Evaluation of Coronary Artery Calcification Score Based on AI chest CT Plain Scan

Objective: To explore the consistency of artificial intelligence-based non-gated low-dose CT(LDCT) plain scanning with ECG gated-CTA on risk stratification of coronary artery calcification score(CACS). Methods: A total of 100 patients with coronary artery CTA scan were analyzed retrospectively. All patients were selected for both coronary ECG gated-CTA scan and routine non-gated chest LDCT plain scan. Agatston calcification score software was applied to record ECG gated-CTA calcification score in Siemens post-processing workstation, and Shukun artificial intelligence analysis software was applied to record non-gated chest CT plain scan calcification score. With the original standard threshold, Kappa test was applied to stratify the risk of non-gated chest LDCT plain scan and ECG gated-CTA assessed CACS. The subject operating characteristic(ROC) curve was adopted to test the diagnostic efficacy of non-gated chest LDCT plain scan and to obtain the optimal new threshold value. The CACS value evaluated by the non-gated chest LDCT with the new threshold value and by ECG gated CTA with old threshold value were selected for further Kappa test. Pearson correlation of calcification integral between the two methods was compared. P<0.001 was set as statistically significant difference. Results: On non-gated chest LDCT plain scan, the risk of CACS was stratified according to the original threshold value, and it was consistent with the ECG-gated CACS value. Kappa value was 0.804 with P<0.001. When comparing the non-gated chest with CACS measured by chest LDCT and that measured by ECG gated-CTA, The area under ROC curve(AUC) of low-medium risk group was 0.910 with P<0.001, and the optimal diagnostic threshold was 112.35. The AUC of the medium-high risk group was 0.988 with P<0.001, and the best diagnostic threshold was 398.31. Consistency test was performed on the risk stratification of CACS measured by non-gated chest LDCT plain scan according to the optimal diagnostic threshold, and the risk stratification of CACS measured with ECG-gated CTA scan according to the original threshold. The Kappa value was 0.850, P<0.001. The Pearson correlation coefficient measured by the two methods was 0.985 and P<0.001, showing a significant correlation. Conclusion: The evaluation of coronary artery calcification by non-gated CT scan based on artificial intelligence and ECG-gated CTA technology is highly consistent. In this study, a new standard for risk stratification of CACS values under non-gated conditions was established, which further improved the accuracy of CACS values evaluated by non-gated chest LDCT and contributed to the early detection of coronary heart disease.