A Novel Dual-scale Dilated Convolutional Neural Network for Heartbeat Classification

Arrhythmia is an important type of cardiovascular disease, which threatens human health greatly. The electrocardiogram (ECG) is used for detecting arrhythmia effectively. However, it is time-consuming for clinician to read ECG recording and detect arrhythmia. Thus, automatic arrhythmia detection system attracts attention recently. A novel dual-scale dilated convolutional neural network (DDCNN) is proposed to classify heartbeat of arrhythmia. Considering comprehensive using ECG information, we combine single heartbeat with corresponding ECG segment as input information. And then, dual-scale input is fed into DDCNN, which includes fully connected layers, dilated convolution layers and pooling layers. Compared with conventional convolutional neural network (CNN), dilated CNN can improve receptive field to have access to a wider range of information. Further, different kernel sizes help convolution kernel to obtain numerous informative features. Finally, the proposed DDCNN is verified by the open dataset and reaches 99.25% accuracy and 93.74% macro F1 score. Compared with other baseline models, DDCNN can achieve good model performance, which demonstrates the effectiveness and superiority of DDCNN.