Detecting severe coronary artery stenosis in T2DM patients with NAFLD using cardiac fat radiomics-based machine learning.
Journal:
Scientific reports
PMID:
40000860
Abstract
To analyze radiomics features of cardiac adipose tissue in individuals with type 2 diabetes (T2DM) and non-alcoholic fatty liver disease (NAFLD), integrating relevant clinical indicators, and employing machine learning techniques to construct a precise model for detecting severe coronary artery stenosis. A retrospective analysis of 710 T2DM patients with NAFLD was conducted at First People's Hospital of Wenling. The study population was randomly divided into a training set (n = 497) and a validation set (n = 213). Radiomics features from cardiac fat CT images, including epicardial adipose tissue (EAT) and paracardial adipose tissue (PAT), were extracted for all patients. The semi-automated segmentation and extraction of shape, first-order statistics, texture, and wavelet were performed using specialized software. Simultaneously, clinical characteristics were collected. Following feature selection, four machine learning algorithms were utilized to develop radiomics, clinical, and combined radiomics-clinical models. The detection performance of these models was subsequently evaluated in both the training and validation cohorts. Additionally, Shapley Additive exPlanations (SHAP) values were calculated to quantify the importance of features. A total of 10 radiomics features for EAT and PAT were extracted from CT images after feature selection. The clinical model obtained an area under the curve (AUC) of 0.747 with the support vector machine (SVM), while the radiomics model reached an AUC of 0.838 with the extreme gradient boosting (XGBoost) algorithm. In comparison, the radiomics-clinical model using XGBoost demonstrated superior detection capability, achieving an AUC of 0.883 in the training set and maintaining high performance in the validation set, with the highest F1 score, accuracy, and precision. SHAP analysis revealed the importance of radiomics features from EAT and PAT, as well as clinical factors such as diabetes duration, global longitudinal strain (GLS), and low-density lipoprotein cholesterol (LDL-C), in detecting severe coronary artery stenosis. This study confirms that the integrated application of cardiac fat radiomics features and clinical data using machine learning models, particularly the XGBoost algorithm, facilitates the detection of severe coronary artery stenosis in T2DM patients with NAFLD. SHAP analysis further elucidates the contribution of key variables in the model, providing crucial foundations for personalized treatment decision-making.