A Short-term ECG Signal Classification Method Based on Residual Network and Bi-directional LSTM

This paper proposes an architecture based on residual network and bi-directional LSTM for analyzing short-term ECG signals. We discriminate basic cardiac disease by using statistical time domain features, frequency domain features, nonlinear domain features and deep features of ECG signals through traditional methods and deep neural networks. Also, this paper integrates the residual network and bi-directional LSTM to focus on the sequence information contained in the ECG signals better. Cinc Challenge 17 database is used as model evaluation. Our method achieves average F1 score of 0.8682 and accuracy of ECG multi-classification 91%, which proves that this method has a certain degree of auxiliary diagnosis of ECG signals.