Radiomics model based on coronary CT angiography for predicting major adverse cardiovascular events in patients with coronary artery disease: comparison of lesion-specific pericoronary adipose tissue model and pericoronary adipose tissue model

TL;DR

This study compares two models using coronary CT scans to predict heart risks in patients with artery disease, finding that a lesion-specific model performs better when combined with clinical data.

Contribution

The study introduces a lesion-specific pericoronary adipose tissue radiomics model that outperforms RCA-based models and improves when combined with clinical features.

Findings

The lesion-specific PCAT model (LS-model) achieved higher AUC values (0.821 and 0.838) compared to the RCA-model (0.789 and 0.788).

The Cli-LS model, combining clinical features and lesion-specific PCAT, achieved the highest AUCs (0.873 and 0.877).

Calibration and decision curve analyses confirmed the superior performance and clinical benefit of the combined models.

Abstract

To assess the performance of a lesion-specific pericoronary adipose tissue (PCAT) radiomics model in comparison to a right coronary artery (RCA) PCAT model in predicting major adverse cardiovascular events (MACE) over a three-year period in patients diagnosed with coronary artery disease (CAD). Additionally, the study aims to evaluate the incremental predictive value of combined models integrating clinical features.

This study conducted a retrospective analysis involving 242 patients with coronary artery disease who underwent coronary CT angiography (CCTA) with MACE occurring in 121 cases. The right coronary artery and lesion-specific PCAT were segmented using the Peri-coronary Adipose Tissue Analysis Tool software (Shukun Technology Co., Ltd.), and 93 radiographic features were extracted, and the features were screened by Pearson correlation coefficients and Lasso regression after the features were processed by Min-Max Normalization. Machine learning techniques were utilized to construct four models: the right coronary artery PCAT model (RCA-model), the lesion-specific PCAT model (LS-model), the clinical model (Cli-model), and two combined models (Cli-RCA model and Cli-LS model). The performance of these models was evaluated by receiver operating characteristic (ROC) curves, calibration curve and decision curve analysis (DCA).

The LS-model demonstrated superior predictive performance with AUC values of 0.821 and 0.838 in the training and test cohorts, respectively. This performance surpassed that ofthe RCA-model, which recorded AUC values of 0.789 and 0.788. Notably, the Cli-LS model achieved the highest AUCs of 0.873 and 0.877. The difference in AUC was statistically significant (p < 0.05). Calibration curves indicated excellent agreement between predicted and observed risks, as indicated by aHosmer-Lemeshow test result of P > 0.05. Furthermore, decision curve analysis confirmed a higher net clinical benefit.

Lesion-specific PCAT radiomics features demonstrate superior predictive capability for MACE compared to f RCA-based features. Integrating clinical risk factors further enhances model performance, offering a noninvasive imaging tool for risk stratification in patients with CAD.

Figures