Machine learning-based radiomic features of perivascular adipose tissue in coronary computed tomography angiography predicting inflammation status around atherosclerotic plaque: a retrospective cohort study.

Abstract

Objectives: This study expolored the relationship between perivascular adipose tissue (PVAT) radiomic features derived from coronary computed tomography angiography (CCTA) and the presence of coronary artery plaques. It aimed to determine whether PVAT radiomic could non-invasively assess vascular inflammation associated with plaque presence.

Methods: In this retrospective cohort study, data from patients undergoing coronary artery examination between May 2021 and December 2022 were analyzed. Demographics, clinical data, plaque location and stenosis severity were recorded. PVAT radiomic features were extracted using PyRadiomics with key features selected using Least Absolute Shrinkage and Selection Operator (LASSO) and recursive feature elimination (RFE) to create a radiomics signature (RadScore).Stepwise logistic regression identified clinical predictors. Predictive models (clinical, radiomics-based and combined) were constructed to differentiate plaque-containing segments from normal ones. The final model was presented as a nomogram and evaluated using calibration curves, ROC analysis and decision curve analysis.

Results: Analysis included 208 coronary segments from 102 patients. The RadScore achieved an Area Under the Curve (AUC) of 0.897 (95% CI: 0.88-0.92) in the training set and 0.717 (95% CI: 0.63-0.81) in the validation set. The combined model (RadScore + Clinic) demonstrated improved performance with an AUC of 0.783 (95% CI: 0.69-0.87) in the validation set and 0.903 (95% CI: 0.83-0.98) in an independent test set. Both RadScore and combined models significantly outperformed the clinical model (p < .001). The nomogram integrating clinical and radiomics features showed robust calibration and discrimination (c-index: 0.825 in training, 0.907 in testing).

Conclusion: CCTA-based PVAT radiomics effectively distinguished coronary artery segments with and without plaques. The combined model and nomogram demostrated clinical utility, offering a novel approach for early diagnosis and risk stratification in coronary heart disease.