A Signal Quality Assessment Method for Electrocardiography Acquired by Mobile Device

Abstract

Electrocardiography (ECG) is a significant tool for detecting cardiovascular diseases. The remote ECG monitoring system by mobile device can gather data anywhere, at any time, which broaden the scope of diagnosis service. However, in clinical, the crucial obstacle involved in the remote system is to identify whether the ECG collected by inexperienced person is usable for diagnostic interpretation. In this study, we address the quality assessment problem of clinical ECG and provide an effective 7-layer Long Short-Term Memory neural network, named LSTM-ECG. According to medical knowledge, we devise a comprehensive feature set which covers the spectral distribution, signal complexity, horizontal and vertical variation of waves, and so on. Meanwhile, we design two LSTM layers in LSTM-ECG to automatically learn the related features. A merge layer is utilized to accomplish feature fusion between domain feature set and LSTM layer feature set and a dropout layer is introduced to prevent overfitting. In order to test the effectiveness of LSTM-ECG, four classifiers are implemented for contrast. Two datasets include large scale clinical data are used in experiments. Comprehensive experiments show that LSTM-ECG is better than the prior state-of-art method and effective in clinical data.