GCN-ResNet: A Multi-label Classifier for ECG Arrhythmia

Automatic ECG classification using artificial intelligence technology is of great significance for the early prevention and diagnosis of cardiovascular diseases. Over the past few years, there have been many studies having demonstrated impressive performance in the area of single label ECG classification. Nonetheless, in real-world clinical scenarios, a single ECG recording may exhibit multiple types of arrhythmia concurrently, thus highlighting the significance of investigating multi-label ECG classification. This paper proposes a multi-label ECG classification model based on GCN-CNN framework, named GCN-ResNet, which can complete the task of multi-label ECG classification by using GCN to extract label information and CNN to extract ECG information. At the same time, two loss function improvement strategies are proposed to address the imbalance problem. Experiments show that our method can achieve F1 score of 0.832 in classifying 55 kinds of arrhythmias, which can help doctors to conduct early rapid ECG diagnosis.