A Machine Learning Model Integrating Tongue Image Features and Myocardial Injury Markers Predicts Major Adverse...

The aim of this retrospective cohort study was to analyse the relationship between markers of myocardial injury, tongue parameters and major adverse cardiovascular events (MACE) in 1293 patients diagnosed with coronary heart disease(CHD).

This was a retrospective cohort study in which data were collected from patients diagnosed with CHD at the Department of Cardiology of Yueyang Hospital of Integrative Medicine and Shuguang Hospital in Shanghai, China, between 1 January 2023 and 31 December 2024, etc. All the patients were classified into two different groups according to follow-up results showed whether there was MACE, and the tongue image of each patient was performed using SMX System 2.0 to normalised acquisition was performed using SMX System 2.0, and tongue body (TC_) and tongue coating (CC_) data were converted to RGB and HSV model parameters. Five supervised machine learning classifiers, including XGBoost, logistic regression, KNN, LightGBM, AdaBoost, were used in building the MACE prediction model.

1293 patients were finally included in this study, with MACE occurred in 279 (21.6%) participants. After sample balancing using the SMOTE method, non-parametric tests revealed significant differences in imaging indicators, some myocardial injury markers, and tongue image parameters between the 2 groups of patients:LDH,MYO,TC_ROOT_R,TC_ROOT_G (P<0.05); the XGBoost, LightGBM models had the highest predictive power (The AUC values of the verification set > 0.97); the combination of SHAP values revealed the importance of the features and provided a quantitative metric to assess the contribution of each feature to the prediction results, Finally, subgroup analysis was conducted based on specific events of MACE.

This study provides insight into the potential application of myocardial injury markers, tongue colour parameters, in the prediction of MACE, and future studies could extend the optimisation of the prediction model and explore its application in other cardiovascular diseases.