Enhancement and optimization of a mobile iBCI for Veterans with paralysis

皮质内脑-计算机接口(IBCI)记录和处理来自 植入大脑皮层的电极阵列，能够快速、准确和直观地控制 为因脊髓损伤、中风而瘫痪的患者提供辅助技术， 或肌萎缩侧索硬化症(ALS)。使用皮质内脑机接口，四肢瘫痪患者 能够使用他们想象中的手部动作来命令在 计算机，使用虚拟键盘打字，使用聊天等通信应用程序，以及浏览 网络。想象中的动作也被用来控制辅助设备，包括DEKA 假臂，辅助机械臂，甚至是自己瘫痪的肢体 对瘫痪肌肉的电刺激。微型无线信号的最新发展 发射器和无线、紧凑、电池供电的神经信号处理器提高了 严重运动障碍患者使用轮椅安装的IBCI的潜力 在没有技术援助的情况下在家里独立工作。要成为一项可行的辅助技术， IBCI不仅必须是移动性的，而且必须是高性能、可靠和直观的。这 研究通过转换算法创新来增强移动IBCI的所有这些方面 在各种临床前研究中展示，并优化它们以实现稳定、高性能 在移动IBCI中进行解码。这项研究首先将高度准确和灵敏的 运行在移动IBCI上的运动学神经解码器(深度学习递归神经网络) 计算能力强大的嵌入式硬件。为了帮助随着时间的推移稳定运动学解码， 提高性能，并简化校准要求，这项研究然后着眼于 本征神经流形使降维(DR)技术适用于 多维、多尺度人类神经数据。接下来，最先进的数据科学方法是 与多类分析集成，以促进对大量数据集的可靠、准确分类 IBCI用户想象的离散手势。接下来，对灾难恢复方法进行评估以解开 同时进行运动学和手势解码，实现更流畅、更准确和不受干扰 IBCI控制中心。这些累积的方法将被转换为嵌入式硬件形式运行 在功能强大的移动处理器上提供按需控制移动和触控 同时使用类似鼠标的移动和手势的设备(如滑动到滚动和捏到 缩放)。将独特的手势映射到其他功能将立即激活快捷键或 手势到短语输出。使用这个安装在轮椅上的IBCI，一个言语残疾的人 可以想象一个手势来生成文本到语音的问候或求救。总体而言，这 研究利用最先进的机器学习创新来实现更有能力、 可靠、多功能的IBCI可促进严重运动障碍患者的独立性。

Intracortical brain-computer interfaces (iBCIs) record and process neural signals streaming from arrays of electrodes implanted in the cortex to enable fast, accurate and intuitive control of assistive technologies for individuals living with paralysis arising from spinal cord injury, stroke, or amyotrophic lateral sclerosis (ALS). Using an intracortical BCI, people with tetraplegia have been able to use their imagined hand movements to command point-and-click actions on a computer, type with a virtual keyboard, use communication apps such as chat, and browse the web. Imagined movements have also been used to control assistive devices including the DEKA prosthetic arm, assistive robotic arms and even one’s own paralyzed limb through patterned electrical stimulation of paralyzed muscles. Recent development of a miniature wireless signal transmitter and a wireless, compact, battery-operated neural signal processor has raised the potential for individuals with severe motor disability to use a wheelchair-mounted iBCI independently at home without technical assistance. To be a viable assistive technology, the iBCI must be not only mobile but also high-performance, reliable, and intuitive to use. This research enhances all of these aspects of a mobile iBCI by translating algorithmic innovations demonstrated in varied pre-clinical studies and optimizing them toward stable, high-performance decoding in a mobile iBCI. This research first transforms a highly accurate and responsive kinematic neural decoder (a deep learning recursive neural network) to run on the mobile iBCI’s computationally powerful embedded hardware. To help stabilize kinematic decoding over time, enhance performance, and ease calibration requirements, this research then looks to theories of intrinsic neural manifolds to adapt dimensionality reduction (DR) techniques to high- dimensional, multiscale human neural data. Next, state-of-the-art data science approaches are integrated with multiclass analyses to promote reliable, accurate classification of a large set of discrete hand gestures imagined by iBCI users. Next, DR methods are evaluated to disentangle simultaneous kinematic and gesture decoding for smoother, more accurate and unperturbed iBCI control. These cumulative approaches will be translated to embedded hardware form to run on the powerful mobile processor to provide on-demand control of mobile and touch-enabled devices using both mouse-like movements and gestures (such as swipe-to-scroll and pinch-to zoom). Mapping unique gestures to additional functions will instantly activate key shortcuts or gesture-to-phrase output. Using this wheelchair-mounted iBCI, a speech-disabled individual could imagine a hand gesture to generate a text-to-speech greeting or call for help. Overall, this research leverages state-of-the-art machine learning innovations toward a more capable, reliable, and versatile iBCI to promote independence for people with severe motor disability.