Zhang Shen, Zheng Yanchun, Wang Daifa, Wang Ling, Ma Jianai, Zhang Jing, Xu Weihao, Li Deyu, Zhang Dan
School of Biological Science and Medical Engineering, Beihang University, Beijing, China.
School of Biological Science and Medical Engineering, Beihang University, Beijing, China; State Key Laboratory of Software Development Environment, Beihang University, Beijing, China; State Key Laboratory of Virtual Reality Technology and System, Beihang University, Beijing, China.
Neurosci Lett. 2017 Aug 10;655:35-40. doi: 10.1016/j.neulet.2017.06.044. Epub 2017 Jun 27.
Motor imagery is one of the most investigated paradigms in the field of brain-computer interfaces (BCIs). The present study explored the feasibility of applying a common spatial pattern (CSP)-based algorithm for a functional near-infrared spectroscopy (fNIRS)-based motor imagery BCI. Ten participants performed kinesthetic imagery of their left- and right-hand movements while 20-channel fNIRS signals were recorded over the motor cortex. The CSP method was implemented to obtain the spatial filters specific for both imagery tasks. The mean, slope, and variance of the CSP filtered signals were taken as features for BCI classification. Results showed that the CSP-based algorithm outperformed two representative channel-wise methods for classifying the two imagery statuses using either data from all channels or averaged data from imagery responsive channels only (oxygenated hemoglobin: CSP-based: 75.3±13.1%; all-channel: 52.3±5.3%; averaged: 64.8±13.2%; deoxygenated hemoglobin: CSP-based: 72.3±13.0%; all-channel: 48.8±8.2%; averaged: 63.3±13.3%). Furthermore, the effectiveness of the CSP method was also observed for the motor execution data to a lesser extent. A partial correlation analysis revealed significant independent contributions from all three types of features, including the often-ignored variance feature. To our knowledge, this is the first study demonstrating the effectiveness of the CSP method for fNIRS-based motor imagery BCIs.
运动想象是脑机接口（BCI）领域中研究最多的范式之一。本研究探讨了将基于共同空间模式（CSP）的算法应用于基于功能近红外光谱（fNIRS）的运动想象BCI的可行性。10名参与者在进行左手和右手运动的动觉想象时，记录了运动皮层上20通道的fNIRS信号。采用CSP方法获得两种想象任务特有的空间滤波器。将CSP滤波信号的均值、斜率和方差作为BCI分类的特征。结果表明，基于CSP的算法在使用所有通道的数据或仅使用想象响应通道的平均数据对两种想象状态进行分类时，优于两种具有代表性的逐通道方法（氧合血红蛋白：基于CSP的方法：75.3±13.1%；所有通道：52.3±5.3%；平均：64.8±13.2%；脱氧血红蛋白：基于CSP的方法：72.3±13.0%；所有通道：48.8±8.2%；平均：63.3±13.3%）。此外，在较小程度上也观察到CSP方法对运动执行数据的有效性。偏相关分析揭示了所有三种类型特征的显著独立贡献，包括经常被忽视的方差特征。据我们所知，这是第一项证明CSP方法对基于fNIRS的运动想象BCI有效的研究。
