摘要
目的基于脑电的运动想象在脑机接口中具有良好的应用前景,本文主要针对运动想象的脑电分类问题,在以往微状态研究的基础上,通过引入LZC、模糊熵、近似熵对微状态序列特征进行计算,判断微状态的序列特征是否可以作为运动想象脑电的神经标志,以及对运动想象分类的影响。方法采用改进的k-means方法对脑电地形图进行聚类,对脑电地形图聚类结果进行返拟,计算脑电微状态地形图参数(持续时间、出现频率、覆盖范围),同时计算脑电地形图的微状态序列特征,包括:Lempel-Ziv复杂度、近似熵和模糊熵,通过统计分析,比较不同任务间的脑电特征差异,进而通过基于高斯内核的支持向量机进行分类。结果不同状态下的脑电信号均获得4个典型的脑电微状态地形图,特征间有显著性差异(P<0.05),且与静息态相比,任务态的微状态D出现频率、覆盖范围较低,任务态的近似熵也较低。通过微状态参数和微状态序列特征融合的分类SVM模型中,对于多种任务两两分类的准确率为90%~98.33%,均高于微状态参数和微状态序列特征的分类准确率。结论将脑电微状态参数和微状态序列特征相融合可以提高脑电分类的准确率以及对于多种任务的分类的泛化能力,且不同状态的脑电微状态序列特征存在显著性差异,故认为脑电微状态序列特征可作为判断运动想象脑电的有效神经生理标志。
Objective EEG-based motor imagery has a good application prospect in brain-computer interface,this thesis mainly focuses on the problem of EEG classification of motor imagery,on the basis of previous microstates research,by introducing LZC,fuzzy entropy and approximate entropy to compute the sequence features of microstates,to judge whether the sequence features of microstates can be used as neural signatures of motor imagery EEG,as well as the influence on the classification of motor imagery.Methods The improved k-means method was used to cluster the EEG topographic maps,and the results of EEG topographic map clustering were back-fitted to calculate the EEG microstate topographic map parameters(duration,frequency of occurrence,and coverage),and at the same time,the microstate sequence features of EEG topographic maps were computed,including:the Lempel-Ziv complexity,the approximate entropy,and the fuzzy entropy,and the statistical analyses were used to compare the different tasks.The differences of EEG features between different tasks were compared by using statistical analysis,and then classified by support vector machine based on Gaussian kernel.Results Four typical EEG microstate topographies were obtained for EEG signals in different states,with significant differences between the features(P<0.05),and the frequency of occurrence of microstate D in the task state,compared with the resting state,had lower coverage.The approximate entropy of the task state was also lower.In the classification SVM model by fusion of microstate parameters and microstate sequence features,the accuracy of two-by-two classification for multiple tasks ranged from 90%to 98.33%,which were higher than the classification accuracies of microstate parameters and microstate sequence features.Conclusions The fusion of EEG microstate parameters and microstate sequence features can improve the accuracy of EEG classification and the generalisation ability for multi-task classification,and there is a significant difference between the EEG microstate sequence features of different states,so it is concluded that EEG microstate sequence features can be used as an effective neurophysiological marker for judging the EEG of motor imagery.
作者
张健聪
王爱坤
张京
ZHANG Jiancong;WANG Aikun;ZHANG Jing(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500)
出处
《北京生物医学工程》
2025年第4期357-364,406,共9页
Beijing Biomedical Engineering
基金
国家自然科学基金(82060329)
云南省基础研究计划项目(02201AT070108)资助。
