Research On Sleep Staging Algorithm Based On Single-lead ECG

As a necessary process of life,sleep is an important part of the body's recovery,integration and consolidation of memory.Sleep staging is not only the basis of sleep research,but also the premise of sleep analysis.It is of great significance for screening and diagnosis of sleep respiratory related diseases to complete sleep staging according to different classification requirements of different scenes.Heart rate variability is closely related to different stages of sleep.In this paper,we study sleep staging algorithm based on single channel ECG.We use ECG signal to extract RR interval,so as to further extract the characteristics of heart rate variability.Sleep staging is carried out under various machine learning and deep learning models,and the classification effect is improved by optimizing data input parameters or parameters in the model,Finally,an optimal sleep staging algorithm is provided for practical verification,and the effect of identifying different sleep types is good,which lays the foundation for the realization of green sleep detection of single channel ECG Device in home environment,and can also be used as a preliminary screening supplement for traditional sleep staging.The main research ideas of this paper are as follows:(1)The single channel ECG data and sleep stage data of SHHS database are used as the basic data for the establishment of early sleep stage model.Three types of sleep stages are realized,namely,the discrimination between awake and sleep stage,the discrimination between light sleep stage and deep sleep stage,the discrimination between awake,rapid eye movement stage and non rapid eye movement stage.(2)RR intervals are extracted from single channel ECG data and preprocessed by upper and lower threshold method and laida criterion.The effective RR intervals are further extracted from time domain,frequency domain and non-linear domain as the original input parameters of each model.(3)The sleep staging models based on support vector machine,random forest,extreme learning machine and long-term memory network were studied.In the parameter optimization of support vector machine,the classification effect of grid search method and genetic algorithm is compared;In the random forest model,the comparison of principal component analysis is added;In the extreme learning machine model,the classification effects of different hidden layers,different types of activation functions and particle swarm optimization are compared;In the long-term and short-term memory network model,the classification effect of unidirectional and bidirectional network is compared,and the bidirectional long-term and short-term memory network model with multi duration heart rate variability as input parameter has the best classification effect.(4)The clinical data were collected in Tianjin Chest Hospital,and the algorithm of the best classification model was selected to predict the sleep stages.Compared with the standard stages modified by professional sleep physicians,the accuracy,sensitivity,specificity,kappa coefficient and chaotic matrix were evaluated,and the feasibility of the algorithm was obtained.Through the research of various sleep staging algorithms,this paper proposes a bi-directional long-term and short-term memory network model based on multi duration heart rate variability.The accuracy rate of discrimination between awake phase and sleep phase is 94.42%,the accuracy rate of discrimination between light sleep phase and deep sleep phase is 92.03%,and the accuracy rate of discrimination among awake phase,rapid eye movement phase and non rapid eye movement phase is84.06%,It provides a reference algorithm for sleep staging based on single channel ECG.