Non-enhanced CT-based radiomics signature of epicardial adipose tissue for screening coronary heart disease

Objective Our study aimed to establish a predictive model based on non-enhanced CT imaging features of epicardial adipose tissue (EAT) to differentiate patients with coronary heart disease (CHD) from those without. Methods In this radiomics study, we collected clinical and radiomic data from a total of 281 patients diagnosed with CHD at the China-Japan Friendship Hospital, along with 188 healthy individuals who underwent physical examinations at our hospital. The participants were allocated to either a training or validation group at random, following a 7:3 ratio. We performed multivariate logistic regression analysis to create a clinical model, using a significance threshold of p < 0.05. Additionally, we employed the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm to highlight important radiomic features for constructing a radiomics model. Lastly, we integrated the clinical and radiomics models to establish a combined model. To assess the model's effectiveness, we used the area under the curve (AUC), DeLong's test, and decision curve analysis (DCA). Results In this radiomics study, the AUC of the clinical model were 0.883 (95% CI: 0.848–0.918) for the training cohort and 0.872 (95% CI: 0.812–0.932) for the validation cohort. In the radiomics model, the AUC for the training cohort was 0.853 (95% CI: 0.814–0.892) and for the validation cohort, it was 0.822 (95% CI: 0.751–0.893). DeLong's test revealed no significant difference in AUC between the clinical and radiomics models in both the training cohort ( p = 0.218) and the validation cohort ( p = 0.24). The combined model exhibited good discriminative ability, and the AUC were 0.930 (95% CI: 0.905–0.956) for the training cohort and 0.914 (95% CI: 0.863 −0.965) for the validation cohort. In the DeLong's test, we found that the AUC of the combined model was significantly higher in both cohorts compared to the other models ( p < 0.05). Furthermore, the DCA curve revealed that using the combined model to identify patients with CHD provided greater advantages compared to using the two separate models. Conclusions Our findings indicate that the combined model, which incorporated clinical features and the radiomics signature of EAT, can serve as a valuable tool for distinguishing patients with and without CHD.