RESUMO
Objective:To explore the predictive value of deep learning (DL)-based coronary artery calcification score (CACS) for obstructive coronary artery disease (CAD) and noncalcified plaque/mixed plaque in type 2 diabetes mellitus (T2DM).Methods:Forty hundred and twenty-four consecutive T2DM patients who accepted CACS scan and coronary CT angiography (CCTA) from December 2012 to December 2019 were included retrospectively, with clinical risk factors and plaque features collected. Plaque composition was classified as calcified, non-calcified or mixed plaque. Obstructive CAD was defined as maximum diameter stenosis≥50%. CACS was calculated with a fully automated method based on DL. Univariate and multivariate logistic regressions were applied to select statistically significant factors and the odds ratios(ORs) were measured. Receiver operating characteristic (ROC) curve was evaluated to assess the predictive performance.Results:Increased CACS was associated with a significantly higher odds of obstructive CAD in CCTA (adjusted ORs were 2.22, 6.18 and 16.98 for CACS=1-99, 100-299, 300-999 vs. CACS=0, and P values were 0.009,<0.001,<0.001 respectively). The area under ROC curve (AUC) of CACS to predict obstructive CAD was 0.764. Compared with 0, increased CACS was associated with increased risk of non-calcified/mixed plaque (adjusted ORs were 2.75, 4.76, 5.29 for CACS=1-99, 100-299, 300-999 respectively and P values were 0.001,<0.001,<0.001 respectively). The AUC of CACS to predict non-calcified/mixed plaque was 0.688. It took 1.17 min to perform automated measurement of CACS based on DL in total, which was significantly less than manual measurement of 1.73 min ( P<0.001). Conclusion:DL-based CACS can predict obstructive CAD and non-calcified plaque/mixed plaque in T2DM, which is economical and efficient, and has important value for clinical diagnosis and treatment.