An automated quantification method for the Agatston coronary artery calcium score on coronary computed tomography angiography.

Background: A coronary artery calcium (CAC) score can provide supplementary information for predicting the risk of cardiovascular disease (CVD). Although CAC is clinically measured with non-contrast cardiac computed tomography (CT), coronary CT angiography (CCTA) may also be used, allowing for the simultaneous evaluation of coronary artery vessels and calcified plaques. This study proposes a method for the automated quantification of the Agatston CAC score from CCTA and compares our method's performance with that of non-contrast cardiac CT. Methods: Sixty-two patients were selected from a clinical registry and divided into four CAC categories. They underwent both non-contrast cardiac CT and CCTA. The Agatston CAC score derived from non-contrast cardiac CT (standard Agatston CAC score) was used as the reference standard. Calcifications were automatically identified and quantified using different thresholds after a deep learning-based coronary artery segmentation model pretrained on CCTA images. Comparisons were made between the standard Agatston CAC score and the CCTA-based Agatston CAC score (CCTA-CAC score) on a per-patient and per-vessel basis. Spearman's rank-order correlation coefficient (R) and intra-class correlation (ICC) values were used to calculate the correlation between the two methods. Results: After comparison, the optimal lower threshold in CCTA-CAC score calculations was found to be 650 Hounsfield units (HU). Using this threshold on a per-patient basis, the automatically computed CCTA-CAC score showed a high correlation (R =0.959; P<0.01) and ICC (R =0.8219; P<0.01) with the standard Agatston CAC score. On a per-vessel basis, the standard Agatston CAC score was also highly correlated with the CCTA-CAC score (R =0.889; P<0.01 and ICC =0.717; P<0.01). Of the 62 patients enrolled, 47 (76%) were classified into the same cardiovascular risk category using the CCTA-CAC score quantification method as when the standard Agatston CAC score was used. Agreement within the CAC categories was also good (kappa =0.7560). Conclusions: Fully automated quantification of the Agatston CAC score on CCTA images is feasible and shows a high correlation with the reference standard. This method could simplify the quantification procedure and has the potential to reduce the radiation dose and save time by eliminating the non-contrast cardiac CT stage.