Multi‐modal machine learning based on electrocardiogram data for prediction of patients with ischemic heart disease

Abstract

Electrocardiogram is a non-invasive method to diagnose patients with cardiovascular diseases. Electrocardiogram analysis methods using deep neural networks have been widely used in recent years and are becoming more efficient and accurate than traditional machine learning methods. Exercise electrocardiogram is useful in the diagnosis of ischemic heart disease, but the study of long-term ischemic heart disease prediction based on exercise electrocardiogram is still limited. To investigate the early predictive value of exercise electrocardiogram in addition to basic physiological data in incident ischemic heart disease in a random general population sample followed from 40 to 69 years of age, a multi-modal fusion model is designed to predict ischemic heart disease based on machine learning methods. With the deep feature extractor and multi-modal feature fusion module, the proposed model performance is better than the traditional expert feature based methods and only deep neural networks. This implies that this proposed model may be improved in a contemporary era of ischemic heart disease prevention with the gradually declining incidence of ischemic heart disease.