Research On Automatic Sleep Staging Method Based On ECG Signal

Adequate sleep is the basis for people to maintain health.Monitoring daily sleep can help people understand their sleep status,and effectively helping people prevent and diagnose sleep-related diseases.The premise of sleep research is sleep staging.The traditional sleep staging is divided manually.However,with the continuous development of science and technology and the improvement of medical standards,the focus of sleep research nowadays is gradually inclined to daily sleep monitoring and automatic sleep staging.Among them,automatic sleep staging using ECG signals which are easy to be collected and processed has become a hot spot of current research.In order to improve the accuracy and reliability of automatic sleep staging method,an automatic sleep staging method based on ECG signals is proposed in this paper,which can be applied to different staging criteria.Two sleep ECG databases(UCDDB and MITBPD)are used for testing.The main work of this paper is as follows:(1)ECG signal preprocessing.Aiming at the problem of inaccurate positioning of the R wave of the ECG signal,this paper uses the GQRS algorithm to locate the R wave,smoothly calibrates it,and eliminates abnormal values to obtain the RR interval and Heart Rate Variability(HRV)signals.This paper constructs the R peak signal from the ECG signal to improve the accuracy of staging.(2)Feature extraction and feature selection of ECG signals.After extracting the time-domain features of the R peak signal and the time-domain,frequency-domain and nonlinear features of the HRV signal,this paper uses the mean to fill the missing values and z-score standardization for the feature set.Then divide the training set and the test set based on stratified sampling,and select the random forest algorithm of recursive feature elimination on the UCDDB training set for feature selection.The simulation results show that the R peak signal proposed in this paper is feasible and effective for sleep staging.(3)Selection and optimization of classification methods.According to the training set of UCDDB,this paper selects the classification method for training and evaluation.The simulation results show that the average accuracy and F₁score obtained using the Gradient Boosting Tree are better than other methods.Then use grid search combined with cross-validation to optimize the hyperparameters of the gradient boosting tree.The simulation results show that the optimized gradient boosting tree can effectively improve the accuracy of sleep staging.(4)Sleep automatic staging method test.To verify the reliability,versatility,and stability of the automatic sleep staging method proposed in this paper.This paper first uses UCDDB and MITBPD to test and compare related literature on sleep staging based on MITBPD.Then test different combinations of features(individual R peak and HRV features)and different populations(based on the apnea-hypopnea index).Finally,sleep efficiency is also predicted.The simulation results show that when the HRV signal is used for sleep staging,adding the R peak signal proposed in this paper can have higher stability and accuracy.Simultaneously,the automatic sleep staging method proposed in this article can be applied to two sleep ECG databases,different staging standards,and different groups of people,with high accuracy,versatility,reliability,and stability.Besides,the automatic sleep staging method proposed in this article can effectively predict sleep efficiency and then analyze insomnia through sleep efficiency.