A Novel Cardiac Arrhythmia Classification Method Using Visibility Graphs and Graph Convolutional Network.

This study introduces an innovative approach for the effective classification of cardiac arrhythmia disease through the integration of the visibility graph and graph convolutional neural network. The initial stage involves preprocessing heartbeats, followed by their conversion into graphs using the visibility graph technique. Subsequently, a diverse set of temporal and graph-based features is extracted for each node within the graph. Finally, a graph convolutional neural network (GCN) is trained on these graphs to classify each heartbeat into its respective arrhythmia class. Electrocardiogram (ECG) recordings from the MIT-BIH arrhythmia database were used to train and evaluate the classifier. The proposed model demonstrated a notable average accuracy of 98.16%, accompanied by an average precision and recall of 98%.