State Key Laboratory of Mechanical Systems and Vibrations, Institute of Robotics, Shanghai Jiao Tong University, Shanghai, People's Republic of China.
These authors have contributed equally to this paper and should be considered as co-first authors.
J Neural Eng. 2020 Sep 11;17(4):046043. doi: 10.1088/1741-2552/ab9987.
Hand movement is a crucial function for humans' daily life. Developing brain-machine interface (BMI) to control a robotic hand by brain signals would help the severely paralyzed people partially regain the functional independence. Previous intracranial electroencephalography (iEEG)-based BMIs towards gesture decoding mostly used neural signals from the primary sensorimotor cortex while ignoring the hand movement related signals from posterior parietal cortex (PPC). Here, we propose combining iEEG recordings from PPC with that from primary sensorimotor cortex to enhance the gesture decoding performance of iEEG-based BMI.
Stereoelectroencephalography (SEEG) signals from 25 epilepsy subjects were recorded when they performed a three-class hand gesture task. Across all 25 subjects, we identified 524, 114 and 221 electrodes from three regions of interest (ROIs), including PPC, postcentral cortex (POC) and precentral cortex (PRC), respectively. Based on the time-varying high gamma power (55-150 Hz) of SEEG signal, both the general activation in the task and the fine selectivity to gestures of each electrode in these ROIs along time was evaluated by the coefficient of determination r . According to the activation along time, we further assessed the first activation time of each ROI. Finally, the decoding accuracy for gestures was obtained by linear support vector machine classifier to comparatively explore if the PPC will assist PRC and POC for gesture decoding.
We find that a majority(L: [Formula: see text] 60%, R: [Formula: see text] 40%) of electrodes in all the three ROIs present significant activation during the task. A large scale temporal activation sequence exists among the ROIs, where PPC activates first, PRC second and POC last. Among the activated electrodes, 15% (PRC), 26% (POC) and 4% (left PPC) of electrodes are significantly selective to gestures. Moreover, decoding accuracy obtained by combining the selective electrodes from three ROIs together is 5%, 3.6%, and 8% higher than that from only PRC and POC when decoding features across, before, and after the movement onset, were used.
This is the first human iEEG study demonstrating that PPC contains neural information about fine hand movement, supporting the role of PPC in hand shape encoding. Combining PPC with primary sensorimotor cortex can provide more information to improve the gesture decoding performance. Our results suggest that PPC could be a rich neural source for iEEG-based BMI. Our findings also demonstrate the early involvement of human PPC in visuomotor task and thus may provide additional implications for further scientific research and BMI applications.
手部运动是人类日常生活的关键功能。通过脑信号开发用于控制机器人手的脑机接口（BMI）将帮助严重瘫痪的人部分恢复功能独立性。以前基于颅内脑电图（iEEG）的手势解码 BMI 大多使用来自初级感觉运动皮层的神经信号，而忽略了来自后顶叶皮层（PPC）的与手部运动相关的信号。在这里，我们提出将来自 PPC 的 iEEG 记录与来自初级感觉运动皮层的记录相结合，以提高基于 iEEG 的 BMI 的手势解码性能。
当 25 名癫痫患者进行三类手部手势任务时，记录立体脑电图（SEEG）信号。在所有 25 名患者中，我们分别从三个感兴趣区域（ROI）中确定了 524、114 和 221 个电极，包括 PPC、后中央皮层（POC）和中央前回（PRC）。基于 SEEG 信号的时变高伽马功率（55-150 Hz），通过决定系数 r 评估每个 ROI 中每个电极的任务中的总体激活以及对随时间变化的手势的精细选择性。根据随时间的激活，我们进一步评估了每个 ROI 的首次激活时间。最后，通过线性支持向量机分类器获得手势的解码精度，以比较探讨 PPC 是否有助于 PRC 和 POC 进行手势解码。
我们发现，所有三个 ROI 中的大多数电极（L：[公式：见文本]60%，R：[公式：见文本]40%）在任务中呈现出显著的激活。ROI 之间存在大规模的时间激活序列，其中 PPC 首先激活，PRC 其次，POC 最后。在激活的电极中，15%（PRC）、26%（POC）和 4%（左 PPC）的电极对运动具有显著选择性。此外，当解码跨、运动前和运动后三个 ROI 的选择性电极时，与仅使用 PRC 和 POC 相比，组合三个 ROI 的选择性电极的解码精度分别提高了 5%、3.6%和 8%。
这是第一项证明人类 iEEG 中 PPC 包含有关手部精细运动的神经信息的研究，支持 PPC 在手部形状编码中的作用。将 PPC 与初级感觉运动皮层相结合可以提供更多信息，从而提高手势解码性能。我们的结果表明，PPC 可能是基于 iEEG 的 BMI 的丰富神经源。我们的研究结果还表明，人类 PPC 早期参与了视觉运动任务，因此可能为进一步的科学研究和 BMI 应用提供了额外的启示。
