Automatic Classification of ECG Signals in WBAN Based on Convolutional Neural Network and Long-Short Term Memory Network

Abstract

To achieve an accurate real-time analysis for the classification of Electrocardiogram (ECG) signals in Wireless Body Area Network (WBAN) has been a critical problem. Hence we propose an automatic classification method of ECG signals based on Convolutional Neural Network (CNN) and Long-Short Term Memory Network (LSTM). The CNN layers are mainly responsible to extract local temporal features and the LSTM layer is mainly responsible to encode the whole temporal history. It can automatically classify the four ECG signals in MIT-BIH Arrhythmia Database. The accuracy, sensitivity and specificity of the classification were 99.86%, 99.86% and 99.96% respectively. For 5289s long ECG beat signals, the above classification effect takes only 4.46s time. In addition, the model has the best classification effect for five types of ECG signals under the AAMI standard. The accuracy, sensitivity and specificity are 99.12%, 99.15% and 99.84%, the classification time is only 0.81s for 756s length ECG signals.