A Novel Convolutional Neural Network for Arrhythmia Detection from 12-lead Electrocardiograms

Abstract

Electrocardiogram (ECG) is a widely medical tool used in the clinical diagnosis of arrhythmia, numerous algorithms based on deep learning have been proposed to achieve automatic arrhythmia detection. In PhysioNetlComputing in Cardiology Challenge 2020, inspired by the deep residual learning and attention mechanism, we proposed a novel neural network to accomplish this classification task. The backbone of the network is a carefully designed 2-D convolutional neural network (CNN) with residual connection and attention mechanism, and it can adapt to multi-lead ECG signals as input. The first 10 seconds of records from all leads are extracted and preprocessed as input for end-to-end training, and the prediction probabilities of 27 categories are output. The proposed algorithm was firstly verified and adjusted via 5-fold cross-validation on officially published datasets from 4 multiple sources. Finally, our team (MetaHeart) achieved a challenge validation score of 0.616 and full test score of 0.370, but were not ranked due to omissions in the submission.