Lee Chungki, Jung Jihee, Kwon Gyuhyun, Kim Laehyun
Korea Institute of Science and Technology, Seoul, 136-791 Korea.
Annu Int Conf IEEE Eng Med Biol Soc. 2012;2012:5290-3. doi: 10.1109/EMBC.2012.6347188.
It is well established that motor action/imagery provokes an event-related desynchronization (ERD) response at specific brain areas with specific frequency ranges, typically the sensory motor rhythm and beta bands. However, there are individual differences in both brain areas and frequency ranges which can be used to identify ERD. This often results in low classification accuracy of ERD, which makes it difficult to implement of BCI application such as the control of external devices and motor rehabilitation. To overcome this problem, an individually optimized solution may be desirable for enhancing the accuracy of detecting motor action/imagery with ERD rather than a global solution for all BCI users. This paper presents a method based on a genetic algorithm to find individually optimized brain areas and frequency ranges for ERD classification. To optimize these two components, we designed a chromosome consisting of 64-bit elements represented by a binary number and another 9-bit elements using 512 pre-defined frequency ranges (2^9). The average value of the significant level is set for the properties of the objective function for use in a t-test, (p < 0.01) depending on the random selection from a concurrent population. As a result, contralateral ERD responses in the spatial domain with individually optimized frequency ranges showed a significant difference between resting and motor action. The ERD responses for motor imagery, on the other hand, led to a bilateral pattern with a narrow frequency band compared to motor action. This study provides the possibility of selecting optimized electrode positions and frequency bands which can lead to high levels of ERD classification accuracy.
运动动作/想象在特定脑区、特定频率范围内会引发事件相关去同步化（ERD）反应，这一点已得到充分证实，通常是在感觉运动节律和β频段。然而，在可用于识别ERD的脑区和频率范围方面存在个体差异。这常常导致ERD的分类准确率较低，使得诸如控制外部设备和运动康复等脑机接口应用难以实施。为克服这一问题，可能需要一种针对个体的优化解决方案来提高利用ERD检测运动动作/想象的准确性，而不是为所有脑机接口用户提供一个通用解决方案。本文提出了一种基于遗传算法的方法，用于找到针对ERD分类的个体优化脑区和频率范围。为了优化这两个组件，我们设计了一种染色体，它由一个用二进制数表示的64位元素和另一个使用512个预定义频率范围（2^9）的9位元素组成。根据从并发群体中随机选择的情况，为目标函数的属性设置显著水平的平均值，用于t检验，（p < 0.01）。结果表明，在空间域中具有个体优化频率范围的对侧ERD反应在静息状态和运动动作之间存在显著差异。另一方面，与运动动作相比，运动想象的ERD反应呈现出双侧模式，且频段较窄。本研究提供了选择优化电极位置和频段的可能性，这可能会带来高水平的ERD分类准确率。
