Evaluation of fully automated commercial software for Agatston calcium scoring on non-ECG-gated low-dose chest CT with different slice thickness.

Objectives: To evaluate commercial deep learning-based software for fully automated coronary artery calcium (CAC) scoring on non-electrocardiogram (ECG)-gated low-dose CT (LDCT) with different slice thicknesses compared with manual ECG-gated calcium-scoring CT (CSCT).

Methods: This retrospective study included 567 patients who underwent both LDCT and CSCT. All LDCT images were reconstructed with a 2.5-mm slice thickness (LDCT), and 453 LDCT scans were reconstructed with a 1.0-mm slice thickness (LDCT). Automated CAC scoring was performed on CSCT (CSCT), LDCT, and LDCT images. The reliability of CSCT, LDCT, and LDCT was compared with manual CSCT scoring (CSCT) using intraclass correlation coefficients (ICCs) and Bland-Altman analysis. Agreement, in CAC severity category, was analyzed using weighted kappa statistics. Diagnostic performance at various Agatston score cutoffs was also calculated.

Results: CSCT, LDCT, and LDCT demonstrated excellent agreement with CSCT (ICC [95% confidence interval, CI]: 1.000 [1.000, 1.000], 0.937 [0.917, 0.952], and 0.955 [0.946, 0.963], respectively). The mean difference with 95% limits of agreement was lower with LDCT than with LDCT (19.94 [95% CI, -244.0, 283.9] vs. 45.26 [-248.2, 338.7]). Regarding CAC severity, LDCT achieved almost perfect agreement, and LDCT achieved substantial agreement (kappa [95% CI]: 0.809 [0.776, 0.838], 0.776 [0.740, 0.809], respectively). Diagnostic performance for detecting Agatston score ≥ 400 was also higher with LDCT than with LDCT (F1 score, 0.929 vs. 0.855).

Conclusions: Fully automated CAC-scoring software with both CSCT and LDCT yielded excellent reliability and agreement with CSCT. LDCT yielded more accurate Agatston scoring than LDCT using fully automated commercial software.

Key Points: • Total Agatston scores and all vessels of CSCT, LDCT, and LDCT demonstrated excellent agreement with CSCT (all ICC > 0.85). • The diagnostic performance for detecting all Agatston score cutoffs was better with LDCT than with LDCT. • This automated software yielded a lower degree of underestimation compared with methods described in previous studies, and the degree of underestimation was lower with LDCT than with LDCT.