Deep convolutional neural networks for electrocardiogram classification

Abstract

With the development of AI, more and more deep learning methods are adopted on medical data for computer-aided diagnosis. In this paper, a 50-layer convolutional neural network (CNN) is trained for normal and abnormal short-duration electrocardiogram (ECG) classification. We do this using a forward neural network with one or more layers of quick connections. This network is deeper than previously used plain network, and it resolves the notorious problem of network degradation of training accuracy and can significantly increase depth to improve accuracy. Detecting fiducial points and combining features are not required, and the classification model can effectively replace the traditional predefined and time-wasting user’s manual selection features. The method was tested on over 150,000 recorded short-duration ECG clinical datasets and achieves 89.43% accuracy, the sensitivity was 87.73%, and the specificity was 91.63%. The experiments demonstrate that our method is efficient and powerful in clinical applications.