Artificial intelligence-assisted examination of cardiac health status through electrocardiogram

Abstract

Abstract Background Cardiac diseases are the main causes of death. Echocardiogram is the standard testing tool of heart diseases but it is inconvenient and expensive. This study aims at developing an artificial intelligence (AI)-assisted method to interpret electrocardiograms (ECG) for cardiac health status. Method A total of 70757 subjects undergoing concomitant ECG and echocardiography were enrolled. ECG was analyzed by using Philips DXL algorithm to obtain 512 parameters, and a healthy heart is defined by normal cardiac chamber size, myocardial thickness, systolic function, diastolic function, and valvular function. A multi-layer perceptron (MLP) and Long short term memory (LSTM) were developed with hold-out validation (training dataset 80% and testing dataset 20%). Three categories of input data are made based on the percentage of healthy data of 30.6%. Death of the participants were confirmed via linking database to the National Death Registry A Kaplan-Meier survival curve analysis was conducted in the testing dataset stratified by AI-determined cardiac health status. Results The testing results of the proposed methods on an independent dataset are shown in Table 1. It is interesting that using MLP has similar performance as the LSTM model. Furthermore, the 30-layer MLP model has a F1 value of 83.8% and the 50-layer MLP model has a F1 value of 83.8, while the F1 value of 2-layer LSTM_0.2 is 85.2% and 2-layer LSTM_0.2 is 84.6%. In addition, AI-determined healthy hearts were associated with better long-term survival. Conclusion An AI-assisted analysis on ECG can act as a preliminary examination of cardiac diseases to facilitate the arrangement of echocardiograms for early detections.