Heartbeat classification using deep residual convolutional neural network from 2-lead electrocardiogram

Background: The electrocardiogram (ECG) has been widely used in the diagnosis of heart disease such as arrhythmia due to its simplicity and non-invasive nature. Arrhythmia can be classified into many types, including lifethreatening and non-life-threatening. Accurate detection of arrhythmic types can effectively prevent heart disease and reduce mortality. Methods: In this study, a novel deep learning method for classification of cardiac arrhythmia according to deep residual network (ResNet) is presented. We developed a 31-layer one-dimensional (1D) residual convolutional neural network. The algorithm includes four residual blocks, each of which consists of three 1D convolution layers, three batch normalization (BP) layers, three rectified linear unit (ReLU) layers, and an identity shortcut connections structure. In addition, we propose to use 2-lead ECG signals in combination with deep learning methods to automatically identify five different types of heartbeats. Results: We have obtained an average accuracy, sensitivity and positive predictivity of 99.06%, 93.21% and 96.76% respectively for single-lead ECG heartbeats. In the 2-lead datasets, the results show that the deep ResNet model has high classification performance, achieving an accuracy of 99.38%, sensitivity of 94.54%, and specificity of 98.14%. Conclusion: The proposed method can be used as an adjunct tool to assist clinicians in their diagnosis. (C) 2019 Elsevier Inc. All rights reserved.