基于脑电信号特征提取的睡眠分期方法研究  

Sleep staging method based on feature extraction of EEG signal

作　　者：刘戈 刘洪运[2] 石金龙[2] 王国静[2] 胡敏露[2] 王卫东[2] LIU Ge;LIU Hong-yun;SHI Jin-long;WANG Guo-jing;HU Min-lu;WANG Wei-dong(Hainan Hospital of Chinese PLA General Hospital,Sanya 572013,Hainan Province,China;Medical Innovation Research Division,Chinese PLA General Hospital,Beijing 100853,China)

机构地区：[1]解放军总医院海南医院,海南三亚572013 [2]解放军总医院医学创新研究部,北京100853

出　　处：《医疗卫生装备》2021年第1期1-8,共8页Chinese Medical Equipment Journal

基　　金：国家自然科学基金资助项目(61701540);国家重点研发计划项目子课题(2016YFC1305703);军队重大科研项目子课题(AWS14R010)。

摘　　要：目的:研究一种基于单通道脑电(electroencephalogram,EEG)信号特征提取的睡眠自动分期方法,发现用于睡眠分期的EEG有效特征,提高睡眠自动分期准确率。方法:使用10名受试者共10925个睡眠EEG样本,提取EEG时域、频域、双谱、非线性共48个特征作睡眠样本的特征向量,构造睡眠自动分期模型。新定义用于睡眠分期的双谱特征,将95%频谱边缘频率、递归定量分析、符号化动力学与熵结合的特征用于睡眠分期(6期)。使用支持向量机(support vector machine,SVM)分类器作睡眠6期自动分类,使用隐马尔可夫模型(hidden Markov model,HMM)修正睡眠分布。结果:基于SVM的睡眠6期的分期准确率达85.06%。应用SVM-HMM睡眠自动分期模型对1名全新受试者整夜睡眠样本作分期,睡眠6期分期准确率为87.34%。结论:该研究提取的48个EEG特征可以作为睡眠分期的有效依据,该睡眠分期方法可用于全新受试者的睡眠自动分期,具有潜在的临床价值。Objective To propose a new kind of single-channel electroencephalogram(EEG)sleep staging method based on feature extraction to find effective features of EEG and improve the accuracy of automatic sleep staging.Methods There were10925 sleep EEG samples from 10 subjects were used to extract a total of 48 features on EEG time domain,frequency domain,bispectrum and nonlinearity for the feature vectors of the sleep samples to construct an automatic sleep staging model.Newly defined bispectral features were applied to sleep staging,and the features combining 95%spectral edge frequency,recursive quantitative analysis,symbolic kinetics and entropy were employed for sleep staging(6 stages).A support vector machine(SVM)classifier was used to automatically classify 6 stages of sleep and a hidden Markov model(HMM)was applied to correcting the sleep distribution.Results The SVM-based classification had the accuracy being 85.06%for 6 stages of sleep;SVM-HMM automatic sleep staging model gained an accuracy of 87.34%for the 6 stages of sleep of some new subject.Conclusion The 48 EEG features extracted in this study can be a valid basis for sleep staging,and the SVMHMM automatic sleep staging method can be used for automatic staging of new subjects,which has potential clinical value.[Chinese Medical Equipment Journal,2021,42(1):1-8]

关 键 词：睡眠自动分期 脑电 双谱 符号动力学 隐马尔可夫模型 

分 类 号：R318[医药卫生—生物医学工程] R338.63[医药卫生—基础医学]