Sleep Stages classification based on electrooculogram and electromyogram toward manned spaceflight
Objective In order to simplify the hygiene processing and reduce the load of sleep monitoring,a method of sleep quality assessment on orbit without EEG is explored.Methods Using the open database ISRUC-Sleep with AASM standard,the training set(n=20)and the test set(sleep disorder group(n=10)and health group(n=10))are completely independent.The electrooculogram(EOG)features include the energy,the root mean square,correlation coefficients and phase-locked values between different frequency bands of two-channel EOG.The electromyogram(EMG)features include fractal dimension,root mean square,the mean value,the maximum value and the root mean square of EMG envelope.Linear support vector machine(LSVM)and random forest(RF)were used to classify wakefulness,REM sleep,light sleep and deep sleep.The accuracy was compared with the results that derived from six-channel electroencephalogram(EEG),two-channel EOG and one-channel EMG.Results Using 50 normalized features of EOG(44 features)and EMG(6 features),for sleep disorder group,kappa coefficients were both 0.75 by RF and by LSVM;for healthy group,the kappa coefficients were 0.73 by RF and 0.70 by LSVM.As a reference for AASM standard,using 140 normalized features of EEG(90 features),EOG(44 features)and EMG(6 features),for sleep disorder group,kappa coefficients were 0.78 by RF and 0.79 by LSVM;for healthy group,kappa coefficients were 0.74 by RF and 0.76 by LSVM.Conclusion The accuracy of sleep scoring from two-channel EOG and one-channel EMG is comparable with that of the gold standard,and can be applied to evaluate the sleep quality during manned spaceflight.
NETL
NSTL
万方数据
