Fully Implantable and Programmable Spike-based Codecs for Neuroprosthetics

描述（由申请人提供）：全世界有数百万人患有神经损伤和疾病，导致严重的运动障碍。通常情况下，残疾是如此严重，以至于不可能养活自己或随时沟通。在过去的十年里，一种被称为脑机接口（bmi）的新型医疗系统已经从研究实验室中出现，现在已经准备好显著改善这些患者的生活质量。bmi直接从大脑的运动结构中“读出”神经电活动，并解码这些电脉冲，以确定预期的运动。控制电脑光标的BMI系统的初始版本目前正在FDA的i期临床试验中，许多机构正在积极参与临床转化（例如，NIH， DARPA， VA）。创造控制信号，使截肢者能够用假肢（机械手臂）和手喂养自己，将需要解码来自数千个电极的信号，而不是目前系统读取的数百个左右的信号，以及将来自手臂和手的数千个传感器信号编码为数千个人工神经信号，并“写入”大脑，这还没有尝试过。缺乏运行bmi编码和解码算法（称为编解码器）所需的低功耗（以便植入）电子电路是成功临床翻译的根本障碍。目前可用的技术要么过于耗电（数字），要么过于算法不灵活（模拟），无法应对这一挑战。神经形态工程的最新进展使制造完全可植入和可编程的编解码器芯片成为可能。这种创新的方法结合了数字和模拟的最佳特性——可编程性和效率——同时提供了比两者都大得多的稳健性。与此同时，神经科学技术的最新进展使得现在有可能获得设计正确算法所需的知识，以便在我们的编解码器芯片上运行。光遗传刺激现在可以用来驱动猕猴皮层的神经元，计算机视觉现在可以用来跟踪自由运动的猴子，同时进行无线记录。我们建议利用这些最新进展，通过以下原则设计来显著提高假肢的性能：(1)一种全新的编码器，可以通过光遗传技术对神经活动进行时空模式。(2)一种全新的解码器，可以在现实世界中使用，动物可以在更少限制的环境中自由移动。(3)一种全新的可植入可编程电子设备，它达到了运行这些复杂算法所需的能源效率水平。我们将通过在一只自由移动的灵长类动物的前运动皮层和体感觉皮层分别植入96微电极记录阵列和9通道光遗传刺激器来证明我们的成功，控制一个类似人类的机械手臂。我们的最终目标是实现神经形态工程师的梦想：用像大脑一样工作的芯片取代受损的神经组织，帮助数以百万计的神经损伤患者。

DESCRIPTION (provided by applicant): Millions of people worldwide suffer from neurological injury and disease resulting in profound movement impairment. Often the disability is so severe that it is not possible to feed oneself or readily communicate. A new class of medical system termed brain-machine interfaces (BMIs) has emerged from research labs in the past decade and is now poised to dramatically improve these patients' quality of life. BMIs "read out" neural electrical activity directly from motor structures in the brain and decode these electrical impulses in order to determine the intended movement. Initial versions of BMI systems that control a computer cursor are now in FDA Phase-I clinical trials, and numerous agencies are actively engaged in clinical translation (e.g., NIH, DARPA, VA). Creating control signals to enable an amputee to feed himself with a prosthetic (robotic) arm and hand will require decoding signals from thousands of electrodes, rather than the hundred or so signals current systems read, as well as encoding thousands of sensor signals from the arm and hand into thousands of artificial neural signals to be "written into" the brain, which has not yet been attempted. The lack of low-power (so that it can be implanted) electronic circuitry needed to run BMIs' encoding and decoding algorithms (termed codecs) is a fundamental barrier to successful clinical translation. The technologies available until now are too power-hungry (digital) or too algorithmically inflexible (analog) to meet the challenge. Recent advances in neuromorphic engineering make it now possible to build a fully implantable and programmable codec chip. This innovative approach combines digital's and analog's best features-programmability and efficiency-while offering far greater robustness than either. Meanwhile recent advances in neuroscience techniques make it now possible to obtain the knowledge needed to design the right algorithms to run on our codec chip. Optogenetic stimulaton can now be used to drive neurons in macaque cortex and computer vision can now be used to track freely moving monkeys while recording wirelessly. We propose to leverage these recent advances to dramatically increase prosthetic performance through the principled design of: (1) An entirely new class of encoders that can spatio-temporally pattern neural activity via optogenetic techniques. (2) An entirely new class of decoders that can operate in the real world with animals moving freely around in far less constrained settings. (3) An entirely new class of implantable programmable electronics that achieves the level of energy- efficiency required to run these complex algorithms. We will demonstrate our success by having a freely moving primate, with a 96-microelectrode recording array and a 9-channel optogenetic stimulator implanted in its premotor and somatosensory cortex, respectively, control a human-like robotic arm. Our ultimate goal is to realize the neuromorphic engineer's dream: Helping untold millions with neurological injury by replacing damaged neural tissue with chips that work like the brain.