Motor neural dynamics of free behavior enabled through 3D computer vision

运动系统神经科学试图理解自愿运动背后的神经机制。最后两个 几十年来，随着多电极记录和统计估计的使用，这一领域发生了变化 和建模技术。这些技术进步产生了丰富的低维神经动力学， 暗示了行为背后的机制。为了尽量减少混淆，这些研究中的绝大多数 利用行为约束将感兴趣的行为隔离出来进行研究。虽然有效地生成了许多行为- 在医学上相似的试验中，这可能会产生无意的结果，即将神经动力学限制在一个子集上 它的全部范围。 这个项目试图更好地了解神经动力学是否以及如何随着行为环境的变化而变化 (受约束与不受约束)它们出现在。这种类型的工作历来具有挑战性，因为捕捉 不受约束的设置中的肢体运动学不是普通的。然而，随着计算机视觉技术的最新进展， 精确的3D相机已经成为人们可以使用的研究工具。这项研究将利用这些新的3D摄像机来捕获 大型观测围栏中的不受约束的行为。处理这些3D数据集的新算法将 用来估计受试者的姿势。这些肢体运动学将与神经数据同步和关联 记录自植入皮质运动区的一个或多个96通道犹他州电极阵列(S)。 可以从这些同步数据中产生低维神经动力学。这一动态将在 围栏中的两种行为的背景：在围栏上行走和伸手觅食。动力学的维度 在这两种情况下，将进行比较，零假设表明在维度上没有差异 这些行为背景之间的动态变化。随后的实验将再次构建低维神经 动力学，但这一次包括了行为约束的到达的背景。来自这三个背景下的动态 将被比作？和所有人共享的公共子空间(维度的子集)。此子空间，如果它存在( 无效假设是行为上下文之间的动态没有差异)，代表基本 与行为背景不变的动力学，暗示着这个子空间的因果必然性。 综上所述，这些研究将进一步加深我们对低维神经动力学驱动电机的理解- 哈沃。这一见解对动态脑机接口的发展具有影响，并可能为 对中风等运动障碍患者的治疗。

Motor systems neuroscience seeks to understand the neural mechanisms behind voluntary movement. The last two decades have witnessed a transformation in this ?eld with the use of multielectrode recordings and statistical estimation and modeling techniques. These technological advances have yielded rich, low-dimensional neural dynamics that are suggestive of the mechanisms underlying behavior. To minimize confounds, the overwhelming majority of these studies utilize behavioral constraint to isolate just the behaviors of interest for study. While effective for generating many behav- iorally similar trials, this may have the unintentional consequence of arti?cially constraining neural dynamics to a subset of its full range. This project seeks to better understand whether and how neural dynamics change with respect to the behavioral context (constrained vs unconstrained) they occur in. This type of work has historically been challenging because capturing limb kinematics in an unconstrained setting is non-trivial. However, with recent advances in computer vision technology, accurate 3D cameras have become accessible tools for research. This study will leverage these new 3D cameras to capture unconstrained behavior in a large observational enclosure. Novel algorithms for the processing of these 3D datasets will be used to estimate the subject's pose. These limb kinematics will be synchronized and correlated against neural data recorded from one or more 96-channel Utah electrode array(s) implanted in motor regions of cortex. Low-dimensional neural dynamics can be generated from this synchronized data. The dynamics will be explored in the context of two behaviors in the enclosure: walking and reaching for food on the ?oor. The dimensionality of the dynamics in these two contexts will be compared, with the null hypothesis stating that there is no difference in dimensionality of dynamics between these behavioral contexts. A subsequent experiment will be to again construct low-dimensional neural dynamics, but this time include a context of behaviorally constrained reaching. The dynamics from these three contexts will be compared to ?nd a common subspace (subset of dimensions) shared among all. This subspace, if it exists (the null hypothesis is that there is no difference in the dynamics between the behavioral contexts), represents fundamental dynamics that are invariant of the behavioral context, suggestive of causal necessity of this subspace. Taken together, these studies will further our understanding of how low-dimensional neural dynamics drive motor be- havior. This insight has implications for the development of ambulatory brain-machine interfaces and may inform the treatment of individuals with motor disorders such as stroke.