【作者】 朱家富；

【导师】 何为；

【作者基本信息】 重庆大学 ， 电气工程， 2009， 博士

【摘要】 大脑是人体最复杂的生理器官,其不但支配人的思维和行为,而且也是控制情绪和植物神经功能的最高神经中枢,这些神经活动对外反映在大脑皮层脑电信号的变化上,对于颅脑损伤所引起的不同程度的意识与认知障碍,也反应在脑电信号的特征参数的变化上。本论文研究的目的是期望以正常人和颅脑损伤病理病人样本的思维脑电信号为研究对象,以相关维值和复杂度两个非线性动力学指标为特征指标,探索正常样本和颅脑损伤病理样本脑电信号的非线性动力学指标在数值上的差异,探索病理样本在不同治疗与康复时期动力学指标的变化趋势与规律,为人体大脑的意识与认知功能评价、脑损伤诊断、损伤康复监测提供临床上初步的数字化参考指标。本研究以实验采集的各种样本在多个不同思维与意识状态下的脑电信号为分析对象,以非线性动力学为理论基础,以信号处理的理论与方法为手段,系统研究了动力学系统指标的计算方法、脑电信号的预处理方法、脑电信号的非线性指标的计算方法、样本数据的计算与分析等。本论文具体完成的研究工作主要包括:(1)深入研究了非线性动力学的理论,讨论了相关维值与Kolmogorov复杂性等非线性特征指标的特性与计算方法,并对相空间的重构参数和指标的各种计算参数的选择作了对比分析,给出了各个参数的参考确定标准,并编制了各指标的完整计算程序,实现了各动力学指标的计算。(2)从吸引子重构、散点图、功率谱特征和数值识别几方面论证了脑电时间序列的混沌特征,结果表明脑电信号具有典型的非线性特征,可以利用非线性指标来描述脑电信号。(3)设计了脑电信号采集时受试者思维与意识任务方案,并采集了不同样本组的临床原始脑电信号。利用数字滤波和小波变换实现了对脑电时间序列中的工频干扰、基线漂移、肌电干扰、心电干扰和眼动干扰等干扰信号的去除。利用思维脑电的阵发性理论实现了脑电信号与思维意识活动之间的同步分离,实现了对脑电信号的有效选段,从而获得可供实验分析的有效脑电数据,改善了分析结果的可靠性与正确性。(4)从相关维值D2计算过程中重构相空间点的高密度性入手,改进了传统G-P算法,大大提高了相关维值的计算速度,同时保证了计算结果的稳定性,并对原始算法及其改进算法的计算工作量进行推导和实验验证。提出了利用高阶复杂度来描述动力学系统特征的复杂性指标,有效降低了对时间序列字符化时信息的丢失,提高复杂度描述的精确性。(5)计算了不同样本在不同状态下的临床脑电数据的相关维值D2和复杂度指标C k,并对计算结果作了不同状态、不同病理和大脑不同区域的对比分析,结果表明相关维值D2和复杂度C k在不同性质的样本、大脑不同区域与不同状态下均存在明显的数值上的差异。在本论文涉及的研究工作中,主要的创新工作包括:(1)提出了利用脑电信号的非线性动力学参数来实现对颅脑损伤病人意识障碍程度的定量化评价方法,此方法与传统的评分评价方法相比,避免了评价需要丰富的临床医学知识和主观性强的缺点,并具有无创性、数值化与可操作性强等特点。(2)设计了脑电信号采集时受试者思维与意识任务的整套方案。在试验组中,根据初步的研究设计了五种实验任务,而在实验组中,设计了八种不同的思维与适宜任务。并在不同的思维与意识任务下,收集了所有的研究用原始脑电数据。(3)利用多种信号处理的方法与手段,剔除了脑电信号中的各种干扰噪声与伪迹,并实现了思维脑电信号的同步分离,提高了脑电信号分析与计算时的有效性,并利用重构相空间点的高密度性对脑电信号的相关维值的计算算法作了改进,在选择合适的时间间隔τ时,既提高了运算速度,又可以保证计算结果的稳定性。(4)给出了不同样本在不同状态下脑电信号的相关维值D2和复杂度C k的分析结果。发现相关维值D2和复杂度C k在描述脑电信号的动力学特征时具有相似的特征表现:人体大脑在进行复杂的思维与意识活动时,相关维值D2和复杂度Ck的数值比大脑在进行简单思维与意识活动时要大;正常人样本的相关维值D2和复杂度C k的数值比同状态下病人样本的数值大;人体大脑的左右半球在相关维值D2和复杂度C k的数值统计上也存在差异;颅脑损伤康复过程中,病人脑电信号的相关维值D2和复杂度C k的数值也发生变化。这一结果在临床上用于对人体大脑功能的评价、大脑疾病与损伤的诊断、大脑疾病与损伤的康复监测等方面都具有重要的临床参考价值。更多还原

【Abstract】 Cerebrum is the most complicated physiological organ of the human body. It dominates people’s thoughts and behaviors, and it is the nerve center of the human body to control emotions and functions of vegetative nerve. Those nervous activities are reflected in the changes of the electroencephalogram(EEG) signals which can be collected from cerebral cortex by electrodes. The degrees of congnition and consciousness disturbance caursed by craniocerebral injury can be described by the values of the parameters of the EEG. In this study, the EEG signals of normal and patient groups in different states are collected as original data, and the nonlinear dynamical parameters, correlation dimension D2 and complexity measure C k, are selected to describe the EEG signals. The aim of this study is to seek the numerical differences of the nonlinear dynamical parameters between the normal group and patient group, is to search the changing pattern of the parameters in different treatment and rehabilition period of patient. The results can be used as a reference to evaluate the cerebral congnition and consciousness disturbance functions, to diagnose the cerebral damages and diseases, and to monitor the healing process of craniocerebral injury in clinics.In this study, the EEG signals of all groups collected in different states are as study subjects, the nonlinear dynamics is as the basic theory, and the theories and methods of signal processing are as the basic means. The computing methods of the nonlinear parameters, preprocessing methods of the EEG signal, computing methods of the EEG signal and the analyzing methods of the data are systematically studied. The main contents of this paper are follows:(1) Based on the studies of the nonlinear dynamics, the characteristics and computing methods of the correlation dimension and the Kolmogorov complexity measures are discussed. After the selection methods of the reconstruction parameters of phase-space and the computation parameters of nonlinear indexes are compared and analyzed, the reference standards to select those parameters are presented. At last, the complete programs are designed to calculate the nonlinear indexes.(2) The chaotic characteristics of the EEG signals are certified by many methods, such as the reconstruction of its strange attractor, scattered plots, power spectrum and numerical recognition. All indicate that the EEG signal is nonlinear signal and it can be described by the nonlinear indexes.(3) The schemes of thought model are designed. All original EEG signals are collected in the schemes. By the digital filtering and wavelet transforming, the EEG noises, such as power-frequency interference, baseline drift, EMG, ECG and EOG artifacts, are removed from the original EEG signals. Based on the thought burst theories, the thought EEG signals are synchronously separated. By the effective selections of the EEG signals, the experiment data are obtained. This process improves the reliability and accuracy of the experiment results.(4) In the calculating process of the correlation dimension D2 , because of the high density of the reconstructed phase-space, the traditional G-P algorithm is improved. By the modified algorithms, the computing speeds of the correlation dimension D2 are increased and the stabilities of the computing results are maintained. The modified algorithms are deduced from the original G-P algorithm and verified by numerical simulations. The high-order complexity measure is introduced to represent the complexity of the EEG signals. By this means, the missing information in the charactering process of the EEG sequences is reduced and the accuracy of the complexity measure to represent the EEG signals is increased.(5) The correlation dimension D2 and complexity measure C k of all samples in different states are calculated. The computing results are analyzed and compared in different thought states and cerebral areas. According to the results, the correlation dimension D2 and complexity measure C k for different samples and states can be remarkably differentiated.The main original contributions of this paper are as follows:(1) The nonlinear dynamic parameters of EEG signals are ulitilized to describe the degrees of disturbance of consiousness for patients in craniocerebral injury. Comparing with the traditional evaluation method which is subjective because it must be fulfilled by clinical doctor, the method is objective, noninvasive, numerical and operable.(2) The schemes of thought model for the EEG signals collection are designed. For test group, 5 thought tasks are designed and 8 thought tasks are designed for experiment groups.(3) By the methods and tools of signal processing, the EEG noises and artifacts are removed from the original EEG signals. Based on the thought burst theories, the thought EEG signals are synchronously separated. By those processes, the reliability and accuracy of the experiment results are improved. According to the high density of the reconstructed phase-space, the traditional G-P algorithm is improved. After the limited interval timeτis selected, the stabilities of the computing results can be maintained.(4) The statistic results of the correlation dimension D2 and complexity measure C k of the EEG signals in different groups and states. By the results, the statistic characteristics of the correlation dimension D2 are similar with the characteristics of the complexity measure C k. The values of the correlation dimension D2 and complexity measure C k are large in the states of complex thought activities. The values of the correlation dimension D2 and complexity measure C k for normal group are larger for patient group. The statistic results of the correlation dimension D2 and complexity measure C k of the EEG signals for all samples are different between right and left side cerebral areas. The correlation dimension D2 and the complexity measure C k of EEG signals for patients in craniocerebral injury are changed with the processes of the treatment and rehabilition. The results can be preliminarily used as a reference to evaluate the cerebral functions, to diagnose the cerebral damages and diseases, and to monitor the healing process of encephalopathy in clinics.更多还原