基于脑电信号的睡眠分期算法研究

Study on Sleep Staging Algorithm Based on EEG Signals

睡眠质量与人类健康息息相关,准确的睡眠质量监测对于帮助人们改善睡眠质量能够起到有效的监督作用。以MIT-BIH多导睡眠数据库slp01、slp02和slp04等3个样本的脑电信号为分析对象,采用sym7小波对其进行7层分解以去除高频细节信号,得到较为纯净的脑电信号。然后通过非线性符号动力学分析,去趋势波动分析以及频谱分析,分别提取符号熵指数,去趋势波动指数以及δ频带能量比等3个参数,对每个样本采用Kennard-Stone方法按照4∶1的比例建立校正集样本和预测集样本,并结合最小二乘支持向量机分类器进行样本训练拟合与分类识别。结果表明,3个特征参数与睡眠状态具有高度相关性,相关系数绝对值均高于0.83,并且确定了符号熵参数的嵌入维数为4,延迟常数为1,去趋势波动指数的分段区间为30~500,平均的睡眠分期正确率可达92.87%,比基于复杂度、近似熵等算法的分类正确率提高约5%。

The quality of sleep is closely related to the human life. Monitoring sleep quality accurately can play an effective role in helping people improve the quality of sleep. We chose the EEG and sleep state data of slp01,slp02 and slp04 samples of MIT-BIH Polysomnographic database as the analysis object,use the wavelet transform of ‘sym7＇with 7 layers decomposition to denoise the EEG signal,and extract the symbolic entropy,the detrended fluctuation index and the delta frequency band energy ratio through the nonlinear analysis of symbolic dynamics,detrended fluctuation analysis and spectrum analysis. Besides,the calibration samples and prediction samples of each sample were established according to the proportion of 4 to 1 by Kennard-Stone method,and the sleep staging are realized by the least squares support vector machine（ LS-SVM）. Results demonstrated that the three parameters were highly correlated to the sleep state,and the correlation coefficients of them to the sleep state were higher than 0. 83,the embedding dimension and time delay of the symbol entropy parameters are 4 and 1,and the interval of detrended fluctuation was 30- 500,the mean of sleep staging accuracy reached 92. 87%. The accuracy improved about 5% compared to the complexity and approximate entropy algorithm.