【作者】 尹刚；

【导师】 尧德中；

【作者基本信息】 电子科技大学 ， 生物医学工程， 2010， 博士

【摘要】 事件相关电位(Event-related potentials简称ERP或ERPs)是和特定事件相关的脑区电位变化,它与一个可定义的参照事件有稳定时间关系,以特定事件出现的时间作为时间参考,事件相关电位可以通过叠加平均得到,这种特定事件可以是刺激或行为反应。如果在单次实验记录中包含两个或多个事件,不同事件引发的ERP将形成相互混叠,以某一事件出现的时间为基准,进行叠加平均提取ERP的时候,总平均ERP中将出现与不同事件引发的ERP成份之间的交叉干扰。ERP高时间分辨率的特性使得它在认知神经科学以及临床医学领域有广泛的应用,但在一些大脑高级认知功能的研究中,这种各事件引发的ERP成份之间的交叉干扰极大地影响了对ERP成份所对应的生理或心理过程的理解。由于这个问题一直以来都没有得到彻底的解决,目前为止,在大脑认知功能的研究中绝大多数研究者依然采用的是传统的叠加平均技术,因而导致了学界对某些认知功能理解的分歧。所以,发展一种准确有效地提取不同事件引发的真实ERP成份对ERP领域研究有重要意义。本文针对多事件ERP中的交叉干扰问题,研究和发展了多事件ERP成份分解方法以及该方法在实际实验数据中的应用。主要内容如下:1.提出了三事件以及多事件ERP成份分解算法。并结合仿真数据讨论了此方法的分解效果。2.针对多事件ERP成份分解算法的特点,系统研究了边界条件和事件时间分布特性对算法的影响。文中通过仿真对比周期边界条件以及非周期边界条件下算法的效果,得到采用周期边界条件明显好于非周期边界条件,能更真实的恢复仿真数据。通过对不同事件时间分布情况下的仿真对比,得出事件时间分布的形状对算法的效果没有影响,而事件时间的标准差对算法的效果有明显影响,事件时间分布的标准差越大,分解效果越好。这部分的研究对分解算法的实际应用以及心理学实验设计有一定的指导作用。3.提出了一种稳健的多事件ERP成份分解方法---基于维纳去卷积的多事件ERP成份分解算法。针对分解算法中固有的病态问题,采用维纳去卷积理论,通过估计信号和噪声的功率谱来确定最优滤波器的设计。通过仿真实验,较系统地对比了不同正则化方法的分解效果,得出维纳去卷积分解方法具有最好的分解效果。针对实际的实验数据,从ERP的基本特性以及基本生理常识对分解结果的可靠性做了分析。4.将维纳去卷积分解方法应用到反应抑制的研究中,并对反应抑制研究中存在分歧的问题做了详细研究,分析结果证实,NoGo-P3效应可能是反应相关成份的干扰导致的,而和反应抑制过程无关,NoGo-N2效应才是和反应抑制过程相关的。更多还原

【Abstract】 Event-related potentials (ERP or ERPs) are the brain electric activities related to specific event; it has a stable time locked relationship with definable reference event and can be estimated by average across many trials aligned to a specific reference event onset time point, i.e. a stimulus or a behavioral response. If two or more events are included in a single trial, the cross-contamination of ERP components evoked by these different events will exist in average ERP when aligned to the reference event onset time. Because of its high temporal resolution, ERP is widely adopted in cognitive neuroscience and clinical medicine field. However, during the research of higher level cognitive function of brain, the cross-contamination of ERP components evoked by different events may greatly influence the understanding of psychology and physiological processes corresponding to ERP components. Since this problem has not been solved, transitional average technology is still applied in many brain cognitive function researches and hence resulting in conflicting conclusions about some cognitive function. Thus, the development of a precise and effective method for extracting ERP components evoked by different events is crucial to ERP researches.In order to eliminate the cross-contamination of ERP components evoked by different events, a multi-events ERP components decomposition method as well as its application in experiments are presented and discussed in this paper.1 Decomposition method of three and more events ERP components is presented. The validity of this method was proved by simulated data.2 Considering the feature of decomposition method, researches into the influence of boundary and event time distribution on the effectiveness of decomposition are carried out. By comparing the effectiveness of algorithms in periodic boundary condition and non-periodic boundary condition, we find that better and more precise result can be obtained using periodic boundary condition rather than using non-periodic boundary condition. Comparing the results among the different event time distribution, we find the shape of event time distribution does not influence decomposition result, while the standard deviation of event time distribution has a significant influence on the decomposition result: better result could be achieved by introducing greater standard deviation of event time distribution. The study of decomposition methods can provide guidance for the design of psychological experiments.3 Multi-events ERP components decomposition algotithm based on Wiener deconvolution is presented to recover event evoked ERP components. To solve the problem of the inherent ill-posed problem of multi-events ERP components decomposition algorithm, we introduced Wiener deconvolution to regularize the ill-posed problem, and then designed the optimal filter to recover event related components by estimating signal and noise power spectral. A comparison of the decomposition results between difference regularization methods shows that Wiener deconvolution decomposition algorithm has the best decomposition effect. When applying it to actual experiment data, we analyzed the reliability of decomposition result based on the features of ERP and physiological common knowledge.4 Wiener deconvolution decomposition algorithm is applied to the researches of response inhibition to verify the long existing disagreements in this field . The preliminary results show that the NoGo-P3 effect of response inhibition comes from response related component that may be not related to response inhibition process, while NoGo-N2 effect is actually related to response inhibition.更多还原