语言任务下脑电时频网络特征提取及在脑机接口中的应用

发布时间：2017-12-27 03:32

本文关键词：语言任务下脑电时频网络特征提取及在脑机接口中的应用　出处：《河北工业大学》2016年博士论文　论文类型：学位论文

【摘要】：探索智能、意识的人脑机理,认识人的行为和情感,创新脑疾病诊断与治疗,以及类脑计算和智能机器人是二十一世纪科学的前沿领域。脑机接口和脑网络研究是该领域的重要研究内容。本文围绕人脑听觉脑电事件相关电位和语言任务信息加工的时频特征分析的科学问题,以脑机接口和脑网络技术为手段,将脑皮层标准电极和微电极相结合,以临床实验室数据为依据,系统研究了听觉认知和语言任务下脑电时频网络特征提取与分类方法,构建了动态的因效脑功能网络,分析了相关时频特征及网络性能,探索了大脑对听觉及语言信息加工的特点、动态处理过程及表征方法,为基于听觉与语言任务的脑机接口和脑网络研究提供了依据,对脑科学和认知神经科学研究具有重要的参考价值。主要的创新性研究工作有:1、在听觉事件相关电位的单次提取中,采用经验模态分解方法,并引入听觉脑电特征信号N2ac,有效地提高了信号分类精度和传输速率。设计听觉oddball刺激范式的脑机接口实验,采用经验模态分解及相关系数进行特征提取,并在传统事件相关电位N200、P300的基础上,引入N2ac信号进行二分类研究。结果表明经验模态分解能够有效提取单次实验事件相关电位;事件相关电位N2ac的引入,可以有效提高分类精度。2、利用脑皮层标准电极和微电极,将时域分析和频域分析相结合,研究高频(High gamma)特征信号在语言任务下脑功能的定位作用。采集被试音节朗读任务时的皮层脑电,比较多频段脑电的时频功率谱,探索了微电极与标准电极的激活状态,并与临床皮层电刺激结果做比较,结果表明发音前后High gamma(70-110Hz)功率显著升高,且可以用于语言区定位,并辅助临床癫痫手术的术前评估。3、对比分析了皮层脑电标准电极和微电极空间分辨率对分类结果的影响。提取被试进行音节朗读任务时,标准电极与微电极的High gamma频段幅值特征,比较标准电极、微电极以及二者结合时的分类正确率,结果发现微电极虽然空间分辨率高,但是电极位置会影响分类结果;两种电极的结合,能够获取全局以及局部脑电特征,可以显著提高分类正确率。4、基于复杂网络构建和分析方法,探索语言加工过程中脑信息流的动态处理过程,并对脑网络性能进行分析和评估。采用多尺度的皮层脑电,构建音节朗读任务时大脑语言区的时变动态有向网络连接,同时采用度中心度和特征向量中心度的方法衡量网络中节点的重要性,进一步辅助临床癫痫手术的术前评估。5、分别采用互相关和时变动态贝叶斯网络分析方法,对比分析网络连接特征在脑机接口中的应用。采用互相关和时变动态贝叶斯网络,分别衡量发音前后脑网络的功能连接与效应连接,进而采用网络连接特征进行分类研究,并与High gamma特征的分类结果进行对比。结果显示多数分类情况下时变动态贝叶斯网络连接系数的分类结果与High gamma的分类结果相一致,时变动态贝叶斯网络连接系数的分类结果要优于互相关系数。
[Abstract]:Exploring the brain mechanism of intelligence and consciousness, recognizing human behaviors and emotions, innovating the diagnosis and treatment of brain diseases, and brain like computing and intelligent robots are the frontiers of Science in the twenty-first Century. The research of brain computer interface and brain network is an important research content in this field. The analysis of time-frequency characteristics of scientific problems on human auditory event-related potentials and language information processing task, the brain computer interface and brain network technology, the cerebral cortex and the combination of standard electrode microelectrode, clinical laboratory data, the system of network frequency feature extraction of auditory cognitive and language tasks EEG and classification method, constructs the dynamic effect due to brain functional network, time-frequency characteristics and network performance analysis, to explore the brain on auditory and language information processing characteristics, dynamic process and characterization methods, provide the basis for the hearing and language tasks in BCI and brain research based on network that has an important reference value for brain science and cognitive neuroscience. The main innovative research works are as follows: 1. In the single extraction of auditory event-related potentials, the empirical mode decomposition method is introduced, and the auditory EEG characteristic signal N2ac is introduced, which effectively improves the classification accuracy and transmission speed of signals. The brain computer interface experiment of auditory oddball stimulation paradigm was designed, and EMD and correlation coefficients were used to extract feature. Based on traditional event related potentials N200 and P300, N2ac signals were introduced to study two classifications. The results show that the empirical mode decomposition can effectively extract the related potential of the single experiment event, and the introduction of event related potential N2ac can effectively improve the classification accuracy. 2, using cortical standard electrodes and microelectrodes, we combine time domain analysis and frequency domain analysis to study the location function of high-frequency (High gamma) characteristic signals in language tasks. Subjects read syllable tasks in the cortical EEG, when more EEG frequency power spectrum, to explore the activation state of microelectrode and standard electrode, and compared with clinical cortical stimulation results, results show that the pronunciation of High before and after gamma (70-110Hz) power increased significantly, and can be used to locate the language areas, evaluation and assist clinical epilepsy surgery before operation. 3. The influence of cortical electroencephalogram (EEG) electrode and microelectrode spatial resolution on the classification results was compared and analyzed. Extraction test syllable aloud task, High gamma frequency amplitude characteristics of standard electrode and the microelectrode, compared to standard microelectrode and the combination of the two electrode, when the correct classification rate, results showed that although the microelectrode of high spatial resolution, but the electrode position will affect the classification results; a combination of two electrodes, can obtain the global and local EEG the characteristics, can significantly improve the rate of correct classification. 4, based on complex network construction and analysis method, we explored the dynamic process of brain information flow in language processing, and analyzed and evaluated the performance of brain network. A multi-scale cortical EEG is used to construct the temporal and dynamic network connectivity of the brain in the syllable reading task. Meanwhile, the importance of the nodes in the network is measured by the method of degree centrality and eigenvector centrality, which further assists the preoperative evaluation of clinical epileptic surgery. 5. Using the mutual correlation and time-varying dynamic Bayesian network analysis method, the application of network connection features in the brain machine interface is compared and analyzed. Cross correlation and time-varying dynamic Bayesian networks are used to measure the functional connectivity and effect connection of brain networks before and after pronunciations, respectively, and then classify them based on the characteristics of network connection, and compare them with the classification results of High gamma features. The results show that the classification results of time-varying dynamic Bayesian network connection coefficients are consistent with those of High gamma classification in most classifications. The classification results of time-varying dynamic Bayesian network connection coefficients are better than those of correlation numbers.
【学位授予单位】：河北工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R318;TN911.7

【相似文献】