The Interpretable Graph Neural Network of ECG Beat Classification

Abstract

Cardiovascular disease is the leading cause of death in the world. Cardiac arrhythmia is detected based on the electrocardiogram (ECG), which is related to cardiovascular disease. The automatic diagnosis of ECG improves efficiency for doctors and assists people who would likely learn some relative knowledge. The Graph Neural Network obtains the correlations between nodes via the node embedding and the graph networks. Therefore, this work develops a novel Graph Neural Network to classify cardiac arrhythmia based on the ECG beat. The ‘BlackBox’ characteristics make some people doubt the trustworthiness of models, so one interpretable method in the models that produce the visual graph is designed. The learning layer of the ECG time series was proposed. In this paper, a visual analysis of the ECG will be conducted, which increases the confidence level of this model. In this article, a visual analysis of the electrocardiogram will be performed, which increases the confidence of the model, and the classification results will be compared with three classic machine learning techniques, with the model achieving a classification accuracy of up to 98.65%. The selected features in the model were also explicitly studied in a visual method.