运动想象的大尺度动态功能网络连接  

Dynamic Large-Scale Functional Network Connectivity in Motor Imagery

作　　者：张涛 江晨阳[2] 李梦晨 尧德中 徐鹏[2] Zhang Tao;Jiang Chenyang;Li Mengchen;Yao Dezhong;Xu Peng(Xihua University School of Science,Chengdu 610039,China;Key Laboratory for Neuroinformation of Ministry of Education,School of Life Science and Technology University of Electronic Science and Technology of China,Chengdu 610054,China)

机构地区：[1]西华大学理学院,成都610039 [2]电子科技大学生命科学与技术学院,神经信息教育部重点实验室,成都610054

出　　处：《中国生物医学工程学报》2019年第4期409-416,共8页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(61522105,81330032);四川省科技厅科技计划项目(2018JY0526);西华大学青年学者后备人才支持计划

摘　　要：运动想象是一个多维度的高级脑认知活动,被广泛应用于脑-机接口控制和临床康复。然而,运动想象应用背后的神经机制仍然不清楚。为进一步理解运动想象潜在的神经机制,从大尺度水平探索运动想象的动态脑网络连接,征集26名健康被试进行运动想象功能磁共振扫描实验。基于运动想象任务态磁共振数据,首先,利用独立成分分析,获取11个大尺度功能子网络,并提取子网络对应的时间序列;然后,利用滑窗分析法,构建动态网络连接矩阵,并对所有的连接矩阵进行k-means聚类分析,得到状态依赖的动态连接;最后,利用网络统计分析方法,评估左/右手运动想象动态网络连接差异。结果表明,机器学习方法能更有效地获取数据特征,得到基于数据驱动的最优窗长为31个时间点,并且对左/右手运动想象的分类准确率达75.6%;运动想象大尺度网络连接模式是一种状态依赖的动态变化过程,共聚类出4个动态重构连接模式;左/右手运动想象大尺度动态网络连接模式的特异性,主要体现在额顶网络(FPN)和背侧注意网络(DAN)与其他子网络之间的交互上。该研究的发现,为理解运动想象潜在的神经机制提供新的观点。Motor imagery(MI)is a multidimensional high-level cognitive ability that is widely applied to brain-computer interface control and clinical rehabilitation.However,the underlying neural mechanisms behind the application of MI are still unclear.To further understand the underlying neural mechanisms of MI,we explored the dynamic large-scale brain network functional connectivity patterns of MI.Twenty-six healthy subjects were recruited for MI functional magnetic resonance scanning experiments.Based on the task fMRI data,first,the independent component analysis was used to obtain eleven large-scale functional sub-networks,and the time series corresponding to the sub-network was extracted.Then,we evaluated the dynamic network connectivity matrixes using the sliding window analysis method.Based on these connectivity matrices,we performed the cluster analysis resulting in state-dependent dynamic connectivity.Finally,the network statistical analysis method was used to evaluate the dynamic network difference in left-hand MI and right-hand MI.Results showed that the machine learning method could obtain data features more effectively,and the optimal window length based on data-driven was 31 time points,and the classification accuracy rate for left/right hand MI was 75.6%.The large-scale network connectivity pattern of MI was a state-dependent dynamic change process,resulting in 4 dynamic reconfiguration patterns.The specificity of large-scale dynamic network connectivity pattern during left-hand/right-hand MI mainly reflected by the interactions between the frontal-parietal network(FPN)and the dorsal attention network(DAN)and other sub-networks.Our findings provided new insights into the underlying neural mechanisms of MI.

关 键 词：运动想象 动态网络 机器学习 滑动窗 

分 类 号：R318[医药卫生—生物医学工程]