Application Of Heart Rate Variability Analysis Based On Hidden Markov Models In Sleep Staging

Sleep status is an important indicator of physical health.The accurate sleep stages can not only help people correctly know their sleep status and quality,but also help doctors diagnose the sleep disorders.Studies have shown there is a correlation between sleep stage and changes in heart rate variability(HRV)indicators.This paper proposed a novel method for automatic sleep staging.By using the recognition ability of hidden Markov model(HMM),the R wave-to-R wave(RR)interval sequence data is extracted from ECG signals.Then,based on the changes in HRV indicators,the overnight sleep states are identified,and the automatic sleep staging is achieved.In this theme,sleep staging based on HRV analysis is studied.The HRV is studied by time domain analysis,frequency domain analysis and nonlinear analysis,and sleep staging is achieved by HMM finally.The research work of this paper is consisted of the following two aspects:(1)In this theme,an adaptive threshold algorithm is used to filter the outliers of the original RR interval sequence data,and then time domain analysis method,frequency domain analysis method and nonlinear analysis method are used to analyze the HRV.When acquiring time domain features,the time domain features are decomposed and reconstructed by the complete ensemble empirical mode decomposition with adaptive noise.The frequency domain analysis uses short-time Fourier transform and wavelet transform respectively to calculate the relative and normalized values of the spectral power for each frequency band,and comparison of the effects of feature extraction of these two methods is given.In addition,nonlinear analysis obtains sleep-related information by calculating sample entropy and detrended fluctuation analysis.(2)There are 31 Hidden Markov Models(HMM)established for 31 features that extracted by time domain analysis,frequency domain analysis and nonlinear analysis.Then,based on the average recognition accuracy of the model,11 optimal features are screened,an independent HMM is established for each feature.Finally,through the idea of integrated voting to determine the final sleep stage and automatic sleep staging is achieved.