睡眠脑电自动分期方法研究

发布时间：2019-02-24 10:56

【摘要】：对睡眠分期进行研究有一定的临床和现实意义,睡眠分期在睡眠质量的评估和睡眠相关疾病的辅助治疗中都有重要的作用。传统的人工睡眠分期有它的局限性：效率低、耗时和耗费人力,因此研究自动睡眠分期具有重要的意义。脑电是分析睡眠最重要的一个生理信号,通过对睡眠脑电信号进行一定的处理,提取能表征不同睡眠期的特征参数,并借助分类器进行睡眠分期。睡眠脑电是一种复杂的、时变的非线性非平稳信号,本文通过结合非线性动力学方法样本熵和时频分析方法希尔伯特黄变换进行睡眠特征的提取。通过计算出每个睡眠片段的边际谱,并在此基础上计算各脑电节律的能量比,并结合睡眠脑电样本熵的特征,把所有这些睡眠特征作为分类器的输入,并借助台湾大学林智仁博士开发的Libsvm分类工具箱来进行睡眠分期。本文所采用的实验数据来源于MIT-BIT的PhysioBank中的Sleep-EDF数据库,选择了10个受试者的两导脑电信号进行睡眠分期的研究。本文主要把睡眠分为觉醒期、NREM 2期.NREM3期(深睡期)、NREM 1/REM(快速眼动期)这几个睡眠期。实验结果表明,通过样本熵和希尔伯特黄变换能有效获取睡眠脑电信号的睡眠特征。不同睡眠期的样本熵值之间存着一定的规律性,在非快速眼动期(NREM),随着睡眠的深入,样本熵值不断减小,在NREM 3、4期达到最小值。利用希尔伯特黄变换求得的脑电信号的边际谱在不同睡眠期具有一定的差异性的,脑电节律的能量比能够很好地表征不同的睡眠期。但是仅仅利用样本熵进行睡眠分期的效果一般,而只利用希尔伯特黄变换方法进行睡眠特征的提取,达到的睡眠分期效果较好。通过结合样本熵和希尔伯特黄变换进行睡眠特征提取,睡眠分期的效果进一步提高,比仅仅用其中一种方法的效果都要好,总体分期准确率达到了89.9%。由此可见,通过结合样本熵和希尔伯特黄变换方法作为特征提取的方法对睡眠进行分期具有比较理想的效果,同时也肯定了用脑电信号进行睡眠分期的可行性。
[Abstract]:The study of sleep staging has some clinical and practical significance. Sleep staging plays an important role in the evaluation of sleep quality and the adjuvant treatment of sleep related diseases. Traditional artificial sleep staging has its limitations: low efficiency, time consuming and labor consumption, so it is of great significance to study automatic sleep staging. EEG is the most important physiological signal to analyze sleep. By processing the sleep EEG, the characteristic parameters of different sleep periods can be extracted, and the sleep stages are carried out with the help of classifier. Sleep EEG is a kind of complex, time-varying nonlinear non-stationary signal. In this paper, the feature of sleep is extracted by combining the nonlinear dynamic method, sample entropy and time-frequency analysis method, Hilbert-Huang transform. By calculating the marginal spectrum of each sleep segment and calculating the energy ratio of each EEG rhythm, and combining with the entropy characteristics of the sleep EEG sample, all these sleep characteristics are used as the input of the classifier. And with the help of Libsvm classification toolbox developed by Dr. Lin Zhiren, University of Taiwan to stage sleep. The experimental data were obtained from the Sleep-EDF database in PhysioBank of MIT-BIT. Ten subjects were selected to study sleep stages with two conductance EEG signals. In this paper, sleep is divided into arousal, NREM 2 and NREM3 (deep sleep), NREM 1/REM). The experimental results show that the sleep characteristics of sleep EEG can be obtained effectively by sample entropy and Hilbert-Huang transform. There was a certain regularity between the entropy values of samples in different sleep periods. In the non-rapid eye movement period, the sample entropy value decreased with the deepening of sleep, and reached the minimum value in the NREM 3 / 4 phase. The marginal spectrum of EEG obtained by Hilbert-Huang transform is different in different sleep periods, and the energy ratio of EEG rhythm can well characterize different sleep periods. But the effect of only using sample entropy to stage sleep is general, but only using Hilbert-Huang transform to extract sleep features, the effect of sleep staging is better. By combining sample entropy and Hilbert-Huang transform to extract sleep features, the effect of sleep staging is further improved, which is better than that of only one of the methods, and the overall accuracy of stage is 89.9. It can be seen that the method of sample entropy and Hilbert-Huang transform is an ideal method for sleep staging, and the feasibility of using EEG for sleep staging is also confirmed.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R740;TN911.7

【相似文献】