Arrhythmia Classification with Attention-Based Res-BiLSTM-Net

Abstract

In the modern clinical diagnosis, the 12-lead electrocardiogram (ECG) signal has proved effective in cardiac arrhythmias classification. However, the manual diagnosis for cardiac arrhythmias is tedious and error-prone through ECG signals. In this work, we propose an end-to-end deep neural network called attention-based Res-BiLSTM-Net for automatic diagnosis of cardiac arrhythmias. Our model is capable of classifying ECG signals with different lengths. The proposed network consists of two parts: the attention-based Resnet and the attention-based BiLSTM. At first, ECG signals are divided into several signal segments with the same length. Then multi-scale features are extracted by our attention-based Resnet through signal segments. Next, these multi-scale features from a same ECG signal are integrated in chronological order. In the end, our attention-based BiLSTM classifies cardiac arrhythmias according to combined features. Our method achieved a good result with an average F1score of 0.8757 on a multi-label arrhythmias classification problem in the First China ECG Intelligent Competition.