PDNet: A convolutional neural network has potential to be deployed on small intelligent devices for arrhythmia diagnosis

Abstract

Heart arrhythmia is a group of irregular heartbeat conditions and is usually detected by electrocardiograms (ECG) signals. Over the past years, deep learning methods have been developed to classify different types of heart arrhythmias through ECG based on computer-aided diagnosis systems (CADs), but these deep learning methods usually cannot trade-off between classification performance and parameters of deep learning methods. To tackle this problem, this work proposes a convolutional neural network (CNN) model named PDNet to recognize different types of heart arrhythmias efficiently. In the PDNet, a convolutional block named PDblock is devised, which is comprised of a pointwise convolutional layer and a depthwise convolutional layer. Furthermore, an improved loss function is utilized to improve the results of heart arrhythmias classification. To verify the proposed CNN model, extensive experiments are conducted on public MIT-BIH ECG databases. The experimental results demonstrate that the proposed PDNet achieves an accuracy of 98.2% accuracy and outperforms state-of-the-art methods about 2%.