PSG Base On Heart Rate Variability

Sleep disorders can seriously affect people's daily life,so early sleep- monitor is of vital importance for the prevention and diagnosis of sleep diseases.At present,the inter-nationally recognized ”Gold-Standard” for sleep staging is polysomnography(PSG)based on EEG,EOG and EMG signals' analysis,which has the problems of complicated op-eration,uncomfortable wearing,equipment expensive and so on.Most of the sleep data is manually staging based on the sleep physician,which is time-consuming and labor-intensive,and is affected by the sleep physician 's experience,so it is easy to cause the consistency of staging results to be unstable.Compared with PSG,non-EEG sleep monitoring devices can get good staging results under the condition of simple operation and high comfort.This paper independently designed and produced a PSG device,which can achievement the collection and recording of sleep data.Meantime,I designed a sleep experiment and collected 103-night EEG,EOG,EMG and ECG signals data,and use the artificial staging results of EEG characteristics as a contrast completed the automatic sleep staging based on HRV.Paper research content is as listed:1.Due to the lack of portable devices for simultaneous acquisition of EEG and ECG signals on the market,this paper designed and produced a PSG device,experiments show that the device can accurately collect and record 3-channel EEG,2-channel EEG,1-channel EMG and 1-channel ECG signals?2.Using the device completed 103 night sleep experiments and collected data? Pro-cessing the data and assisting two physicians to completed sleep staging based on EEG characteristics? Organizing the result of staging,and use the result of staging label as con-trast for HRV sleep staging?3.To meet the different needs of sleep staging in scientific research and wear-able sleep devices,this paper uses the Xgboost algorithm as a framework to construct five automatic sleep staging models of five-classification,three-classification and two-classification which based on different combinations of HRV features,and achieve auto-matic sleep staging.In the five-classification model,the accuracy of model 0 and model 4 reached 79.7%?84.0%,the F1 scores reached 78.7% and 83.2%? In the three-classification model,the ac-curacy of model 0 and model 4 reached 85.5% and 89.1%,the F1 scores reached 85.2%and 88.9%? In the two-classification model,the accuracy of model 0 and model 4 reached93.5% and 95.2%,the F1 scores reached 92.9% and 94.9%.