Sleep Staging Based On ECG Signals And Deep Neural Networks

Sleep is a necessary process of life and it is an important part of body's recovery,integration and consolidation of memory.Sleep staging is the basis of sleep research and also the premise of various sleep analysis.Therefore,it is of great significance for the diagnosis of sleep-related diseases to realize sleep monitoring and sleep quality evaluation under home environment.Related studies have shown that the heart rate variability in sleep presents a similar periodic changes with EEG and ECG signals are closely related to different stages of sleep.Based on this research,a new method of sleep staging is put forward.It uses ECG signal which is simple and easily collected to identify sleep wakefulness,rapid eye movement and non-rapid eye movement,laying the foundation for achieving green sleep monitoring under home environment.The automatic stages of sleep can be realized through the design of the deep neural network model,and the results can meet the requirements of most cases.The content of this paper is as follows:1.This paper analyzes the correlation between heart rate variability and sleep stages during sleep.Time domain correlation analyzes correlation from statistics and frequency domain analyzes correlation from fast Fourier transform.The nonlinear characteristics are mainly analyzed from sample entropy and detrended fluctuation analysis.The correlation between heart rate variability and sleep staging is verified through the analysis of these three surfaces.2.Based on the nonlinear mapping relationship between ECG and sleep staging,we choose the Deep Neural Network DNN sleep stage model which are consisted of Stacked Auto-encoder(SAE)and softmax.In order to enhance the robustness of the network and avoid over-fitting,Denoising Auto-Encoder(DAE)and dropout processing are respectively added to the network.We discuss the parameters of the network and analyze the influence of different parameters on the network.Finally,the optimal parameter combination of the network is determined by comparing the prediction error rates of sleep stages under different parameters.3.The proposed DNN sleep stage model is compared with the Principal Component Analysis(PCA)and Support Vector Machine(SVM)combination model to determine the validity of the DNN sleep stage model.It is confirmed that selecting the sample period of 60s for sleep stage can have better effect than 30s.Finally,the number of sleep stages are compared,and the two,three and five stages of sleep are compared respectively.In this paper,through the determination of DNN sleep staging model,the prediction of sleep staging based on ECG signal is realized,and the accuracy of forecasting the sleep staging is over 80%when the sample period is 60 seconds.