【作者】 欧阳高翔；

【导师】 李小俚；

【作者基本信息】 燕山大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 从20世纪80年代,随着非线性动力学理论的发展,非线性动力学被广泛地用于分析癫痫脑电信号。然而,由于基于混沌理论的非线性动力学分析方法需要假设脑电是低维混沌信号,对脑电信号的数据长度及平稳度要求较高,且对脑电信号中的噪声很敏感。为了克服传统非线性动力学方法的缺点,本文立足于癫痫脑电信号分析的要求,提出了新的方法用于刻画癫痫发作各阶段脑电信号的动力学特征变化。首先,结合传统递归图和排序递归图提出了混杂递归图方法分析失神发作脑电信号。该方法的新颖性在于定义时间序列的递归状态时,不仅考虑局部相空间距离而且考虑局部排序模式分布结构。仿真分析表明,基于混杂递归图的对角线结构分布的确定性测度DET,能够刻画模型参数的变化,并且抗噪声干扰能力更强。该方法用于分析大鼠失神发作间隙期、发作前期和发作期脑电信号的确定性特征,发现发作前期脑电信号的DET均值显著地大于发作间隙期,但显著地小于发作期。结果表明大鼠失神性癫痫发作前期脑电的确定性特性程度比发作间隙期脑电更高,但是比发作期脑电更低。其次,为了进一步调查失神性癫痫发作各阶段脑电信号隐藏的非线性动力学特征,基于脑电信号的排序模式分布特性,提出了一个新颖的相异性指数方法。该方法通过计算排序模式分布的距离来分析两段脑电信号的相异性。由神经元群模型模拟生成脑电信号,仿真分析相异性指数的性能。结合移动窗口技术,该方法用于分析110段大鼠失神发作脑电信号,其中58段失神发作脑电信号在其发作前相异性指数有显著地增加,能够成功地检测到发作前期状态,表明该方法可以用于检测失神发作脑电信号的动力学特征变化。再次,针对脑电同步分析在癫痫发作研究中的重要意义,基于排序模式分析提出了一个新颖的互信息估计方法。该方法通过对时间序列排序模式进行分类,来实现复杂的概率分布估计,从而直接估计出时间序列的信息量。通过耦合Henon映射模型和耦合神经元群模型模拟生成耦合时间序列,仿真分析排序互信息方法的性能。相比于传统的直方图方法,基于排序模式的互信息估计方法能够更好地刻画模型耦合系数的变化,而且抗噪声干扰能力更强。该方法应用于分析癫痫发作脑电信号,发现脑电信号间的互信息随着癫痫发作的开始而逐渐增加,在完全发作时达到最大。最后,提出了排序自互信息法用于刻画癫痫发作间隙期、发作前期和发作期脑电的动力学特征。仿真分析发现噪声和混沌序列的自互信息AMI(δ)值随着时滞偏移δ增大而衰减的速率不同,表明自互信息方法能够区别噪声和混沌序列。该方法用于分析癫痫发作各阶段脑电信号,发现发作间隙期、发作前期和发作期脑电的自互信息值分别在时滞偏移δ大约为5-6、7-8和9-10时达到稳定值。进一步结合线性判别分析,表明排序自互信息能够识别不同发作阶段的脑电信号。更多还原

【Abstract】 Since the 1980s, new measures of epileptic EEG based on discipline of nonlinear dynamical systems (chaos) have been developed in the last decade. However chaos-based approaches need to assume that EEG data possesses a non-evolving low-dimensional attractor, and requires a long, stationary and noiseless EEG data to compute the reconstructed attractor’s properties. To overcome the drawbacks of traditional nonlinear methods and meet the requirement of epileptic EEG analysis, this dissertation develops new methods to characterize EEG changes in different epileptic seizure phases.Firstly, the hybrid recurrence plot (HRP), based on traditional recurrence plot (RP) and order recurrence plot (ORP), is proposed to analyze the absence EEG. The innovation of HRP is that the recurrence is defined not only by local phase space distance, but also by the local order patterns structure of a time series. The simulation results demonstrate that the determinism measure DET, based on the diagonal structure of HRP, can reveal the changes of model parameters; and the HRP method is much more robust against noise than the traditional RP and ORP methods. Furthermore, the HRP is applied to indicate the deterministic dynamics of EEG recordings at the seizure-free, pre-seizure and seizure states in genetic absence epilepsy rat. It is found that the DET values of pre-seizure EEG data are significantly higher than those of seizure-free intervals, but lower than those of seizure intervals. These results demonstrate that EEG epochs during pre-seizure intervals exhibit a higher degree of determinism than seizure-free EEG epochs, but lower than those in seizure EEG epochs in absence epilepsy rat.Secondly, in order to further investigate hidden nonlinear dynamic characteristics in EEG data for differentiating absence seizure phases, this dissertation proposes a novel dissimilarity measure based on the ordinal pattern distributions of EEG recordings. The dissimilarity between two EEG epochs can be qualified via a simple distance measure between the distributions of order patterns. A neural mass model is proposed to simulate EEG data and to valid the performance of the dissimilarity measure. Furthermore, the proposed dissimilarity measure is applied to analyze absence EEG data with moving-window technique, which show that this measure successfully detects pre-seizure phases prior to their onset in 58 out of 110 seizures. This suggested that the dissimilarity measure could be used to detect changes in the dynamics of absence EEG data.Thirdly, because the synchronization analysis of EEG recordings has a great role for the study of epileptic seizures, a novel method, based on order pattern analysis, is proposed to estimate the mutual information between EEGs. The proposed method calculates the probability distribution of time series based on the classification of order patterns to directly estimate the amount of information in time series. The coupled Henon map model and coupled neural mass model are proposed to generate coupled time series and to valid the performance of the proposed mutual information method. Compared to traditional histogram method, this proposed method is better for presenting the change of the coupling coefficient in coupled models, and is more robust to noise. Furthermore, the proposed mutual information estimation is applied to analyze epileptic EEG data, which show that mutual information between EEGs gradually increases until it reaches its maximum at full seizure.Finally, permutation auto mutual information (AMI) is proposed as a tool to evaluate the dynamic characteristics of EEG during seizure-free, pre-seizure and seizure phase, respectively. Simulation results show that the AMI rate of decrease with increasing delay shiftedδis different between noise and chaotic series, which demonstrates that AMI method could be able to distinguish between these two different series. Using this proposed method to analyze epileptic EEG data, the results show that the AMI(δ) gradually decrease and firstly reach to constant value about atδ=5-6,δ=7-8 andδ=9-10 in seizure-free, pre-seizure and seizure phase, respectively. Furthermore, combining the LDA classifier, the results confirm that the AMI method has potential in classifying the epileptic EEG recordings.更多还原