The Use of Artificial Neural Networks to Study the Role of a Complex of Biomarkers During the Atherosclerotic Process of Coronary Arteries in Patients With Coronary Artery Disease

Coronary heart disease (CHD) remains one of the leading causes of mortality worldwide. The strategy of early detection of patients with high risk of coronary artery atherosclerosis using machine learning (ML) methods to develop predictive models of CHD is a promising one. The aim of the study: to investigate the association of serum concentrations of a complex of laboratory markers with the severity of atherosclerotic lesion in patients with CHD by developing and optimizing an ML model. Material and Methods. 220 patients with confirmed CHD (according to coronary angiography with SYNTAX Score calculation) were included in the study. The following biomarkers were evaluated: CRP, TNF-α, ESM-1, sEng, Bcl2, Bax, p53, TRAIL, GDF-15, NRG1. Model development was performed with Statistica software (Statsoft, Inc., USA) using the Automated Neural Networks module. Results. The best prediction accuracy was demonstrated by the multilayer perseptron (MLP) 12-5-3 model (training accuracy – 98.71%, testing accuracy – 93.75%, validation score – 93.75%). Sensitivity analysis showed that GDF-15, p53, Bcl2/Bax, and ESM-1 were the most significant predictors of significant and severe atherosclerosis. Discussion. The results confirm the efficacy of MLP in the diagnosis of coronary artery atherosclerosis, allowing us to take into account nonlinear biomarker interactions. The use of biomarkers specific to key aspects of atherosclerosis pathogenesis allowed us to achieve higher classification accuracy, to estimate the significance of each of the indicators in the context of prognostic accuracy. Conclusion. ML methods, in particular MLP, can be an effective tool for predicting coronary artery atherosclerosis severity. Future research will focus on increasing the amount of data, testing the model on heterogeneous patient cohorts.