Optimizing co-adaptation in motor BCIs by uncovering brain-decoder interactions

通过揭示大脑-解码器相互作用来优化运动脑机接口的共同适应

• 批准号: 10775032
• 负责人: Amy L Orsborn
• 金额: $ 70.12万
• 依托单位: UNIVERSITY OF WASHINGTON
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-25 至 2028-08-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10775032
• 关键词: Algorithms; Animal Model; Behavioral; Brain; Clinical; Complex; Data; Devices; Electrocorticogram; Engineering; Failure; Feedback; Future; Goals; Implant; Knowledge; Lead; Learning; Machine Learning; Maps; Measures; Methods; Monitor; Monkeys; Motor; Motor Activity; Motor Cortex; Movement; Neurons; Outcome; Paralysed; Performance; Persons; Process; Property; Signal Transduction; Specificity; Structure; System; Task Performances; Techniques; Testing; Time; Training; Vision; Weight; brain computer interface; computerized tools; improved; neural; neurophysiology; neurotransmission; next generation; novel; real world application; sensor; tool

Project Summary Brain-computer interfaces (BCIs) hold great promise to restore movement to paralyzed people. But BCIs cannot yet provide reliable performance across the long timespans and varied settings needed for real- world applications. Maintaining robust BCI performance over many days is challenging because brains are highly plastic. Plasticity during extended BCI practice leads to changes in how neural activity relates to move- ments—the brain’s encoding of BCI movement. How the brain’s encoding changes is influenced by the decod- ing algorithm used by the BCI to map neural activity into movement. These interactions create complex dynam- ics where methods that improve performance in the short term may produce problems longer-term. Indeed, our preliminary data suggests current adaptive decoding methods used to maintain performance over time lead the brain to form encoders where very few neural signals control movements, which make BCI vulnerable to cata- strophic failure with loss of a single neural signal. The long-term vision of this proposal is to expand the engi- neering tools available to produce robust, high-performance BCIs by building tools that account for and even leverage plasticity. To enable this vision, this proposal will test the overarching hypothesis that decoder-en- coder interactions can be used to jointly optimize BCI performance and robustness. We focus on robustness of BCI systems to signal loss and changes in task context. We will use an animal model where monkeys move cursors with activity from motor cortices, which has repeatedly informed clinical BCIs. We will leverage novel micro-electrocorticography implants that allow us to longitudinally monitor and manipulate cortical dynamics to advance our understanding of plasticity in multi-day (10 days) BCI training. We will test our overarching hy- pothesis across three aims. If our hypothesis is true, there must be relationships between decoders and prop- erties of encoders that are related to robustness. Aim 1 will determine whether decoders influence how infor- mation is structured in an encoder, which influences how robust BCIs are to signal loss. Aim 2 will determine whether decoders influence the specificity of learned encoders to BCI movements, which influences how ro- bust BCIs are to changes in tasks. Finally, if our hypothesis is true, it requires computational tools that can opti- mize multiple goals in a BCI. Aim 3 will test a novel decoder training paradigm we developed that can consider multiple objectives. We will compare our novel method to established single-objective methods to determine whether multi-objective methods can improve robustness without compromising performance. Across all aims, we will perform offline analyses and online perturbations to measure robustness to signal loss and changes in neural state and behavioral task. Together, these studies will identify how critical plasticity computations can be influenced through the decoder. Pairing this with tests of novel decoding methods will establish new frame- works to build encoder-informed decoders and pave the way for BCIs that can leverage brain plasticity.

项目摘要 脑部计算机界面（BCIS）具有巨大的希望，可以恢复对瘫痪者的运动。但是BCIS 还无法在长时间的班级和实现所需的各种设置中提供可靠的性能 世界应用。在很多天中，保持稳健的BCI表现是具有挑战性的，因为大脑是 高度塑料。扩展BCI实践中的可塑性会导致神经活动与运动的关系变化。 ments-大脑对BCI运动的编码。大脑的编码变化如何受解码的影响 - BCI使用的算法将神经活动映射到运动中。这些相互作用创造了复杂的动态 - 在短期内改善性能的方法可能会长期产生问题。确实，我们的 初步数据表明，当前用于维持随着时间的性能的自适应解码方法领导 大脑形成编码，很少有神经信号控制运动，这使得BCI容易受到cata的影响 曲折衰竭，失去单个神经信号。该提议的长期愿景是扩大工程 NEERing工具可通过构建构建工具，甚至可以说明 利用可塑性。为了实现这一构想，该提议将检验总体假设，即解码器 - 编码器的相互作用可用于共同优化BCI性能和鲁棒性。我们专注于鲁棒性 BCI系统以信号丢失和任务上下文的变化。我们将使用猴子移动的动物模型 有运动皮质活性的光标，该活动已反复通知临床BCIS。我们将利用小说 微色皮质学的直感使我们能够纵向监测和操纵皮质动力学 在多天（10天）的BCI培训中，我们对可塑性的理解提高了我们的理解。我们将测试我们的总体 在三个目标中进行戏essis。如果我们的假设是正确的，则必须在解码器与支撑物之间存在关系 与鲁棒性相关的编码器的兴趣。 AIM 1将确定解码器是否影响信息 构造是在编码器中构造的，这影响了BCIS信号丢失的稳健性。 AIM 2将确定 解码器是否影响学到的编码者对BCI运动的特异性，这影响 BUST BCIS将改变任务。最后，如果我们的假设是正确的，则需要可以选择的计算工具 在BCI中进行多个目标。 AIM 3将测试我们开发的新型解码器训练范式，可以考虑 多个目标。我们将将我们的新方法与已建立的单目标方法进行比较以确定 多目标方法是否可以改善鲁棒性而不会损害性能。在所有目标中， 我们将进行离线分析和在线扰动，以衡量鲁棒性以发出信号丢失和变化 神经状态和行为任务。这些研究将共同​​确定关键可塑性计算如何 受解码器的影响。将其与新型解码方法的测试配对将建立新的框架 - 努力为可以利用大脑可塑性的BCI铺设编码器信息的解码器并为BCI铺平道路。