Development of an EMG-controlled BCI for biomimetic control of hand movement in humans

摘要 当被问到时，大多数患有高度脊髓损伤(SCI)的人会选择脑部手术来改善他们的 手动控制，然而即使是最先进的皮质内脑计算机接口(IBCI)也只能进行有限的控制 手指的运动，并且不能直接控制施加的力量。运动学在IBCI中的独家应用 尽管初级运动皮质(M1)有丰富的运动信息表示，但控制仍然存在。我们建议 通过模仿哺乳动物的神经肌肉系统来解决这些基本限制，哺乳动物的神经肌肉系统控制着 指的运动及其通过调节肌肉活动而施加的力量。我们将开发一个IBCI 从人类的M1记录预测预期的肌肉活动(EMG)，然后使用这些EMG信号来控制 关节运动学，他们的僵硬和抓地力，通过一个向前的肌肉骨骼模型的手。我们 假设这种基于肌电的IBCI将是患有高水平脊髓损伤的人类实现更多目标的直观手段 与现有的运动学iBCI相比，对他们的手部运动的通用控制。 瘫痪人类使用的解码器通常是通过记录来自用户的M1的尖峰活动来构建的 试图模仿观察到的光标或机械臂的运动。解码器是通过相关运算来计算的 用观察到的运动学测量M1活动。虽然在概念上相似，但我们对解码器的方法 开发更具挑战性，因为它所需的高维电机输出信号-肌电-不能 被直接形象化或模仿。为了绕过这个问题，我们将记录肌电(以及手势和 当身体健全的人进行广泛的运动动作时。我们还将记录M1峰值 瘫痪的个体观察并试图模仿同样的动作时的活动。健壮的肌电数据 将提供的输出信号用于解码器计算，类似于使用中观察到的弹道 运动学解码器。这种基于肌电的实时IBCI将允许参与者控制一只手，并使用它来应用 以一种模仿自然运动控制的方式向被抓住的物体施加压力。初始开发将在虚拟环境中完成 现实(VR)。随后，参与者将使用相同的仿生IBCI来控制任务中的机械手 旨在复制日常生活中的活动。我们会将用户的性能与这款仿生IBCI进行比较，以 一台最先进的运动学IBCI。一旦成功，这些方法将应用于控制 用于肢体缺失患者的机器人肢体，并作为通过以下方式恢复使用者自身肢体运动的手段 功能性电刺激。它们也可以应用于腿部，在那里控制相互作用力和 肌肉协同收缩所产生的肢体阻抗也很关键。

Abstract When asked, most persons with high-level spinal cord injury (SCI) would elect brain surgery to improve their hand control, yet even the state-of-the-art intracortical brain computer interfaces (iBCI) have only limited control of finger motion and no direct control of applied forces whatsoever. The exclusive use of kinematics in iBCI control is despite the rich representation of kinetic information in primary motor cortex (M1). We propose to address these fundamental limitations by mimicking the mammalian neuromuscular system, which controls both digit motion and the forces they exert through the modulation of muscle activity. We will develop an iBCI that predicts intended muscle activity (EMG) from M1 recordings in humans, then use these EMG signals to control joint kinematics, their stiffness, and grasp forces, through a forward musculoskeletal model of the hand. We hypothesize that this EMG-based iBCI will be an intuitive means for humans with high-level SCI to achieve more generalizable control of their hand movements than with existing kinematic iBCIs. Decoders for use by paralyzed humans are typically built by recording spiking activity from M1 as the user attempts to imitate the observed motion of a cursor or a robotic arm. The decoder is computed by correlating measured M1 activity with the observed kinematics. Though similar in concept, our approach to decoder development is more challenging, as the high-dimensional motor output signals it requires – the EMGs – cannot be directly visualized or imitated. To circumvent this problem, we will record EMGs (as well as hand posture and contact forces) as able-bodied people perform a broad range of motor actions. We will also record M1 spiking activity as paralyzed individuals observe and attempt to imitate the same actions. The able-bodied EMG data will provide the output signals for decoder calculation, analogous to the use of observed the trajectory in kinematic decoders. This real-time, EMG-based iBCI will allow participants to control a hand, using it to apply forces to grasped objects in a way that mimics natural motor control. Initial development will be done in virtual reality (VR). Subsequently, participants will use the same biomimetic iBCI to control a robotic hand in tasks designed to replicate activities of daily life. We will compare the users' performance with this biomimetic iBCI to that of a state-of-the-art kinematic iBCI. When successful, these methods will have application to the control of robotic limbs for patients with limb loss, and as a means to restore movement of the user's own limbs through Functional Electrical Stimulation. They could also be applied to the legs, where control of interaction forces and limb impedance through muscle cocontraction is also critical.