Hierarchical cortical circuits implementing robust 3D visual perception

分层皮质电路实现强大的 3D 视觉感知

• 批准号: 9769032
• 负责人: Ari Rosenberg
• 金额: $ 42.03万
• 依托单位: UNIVERSITY OF WISCONSIN-MADISON
• 依托单位国家: 美国
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-09-01 至 2023-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9769032
• 关键词: 3-Dimensional; Animals; Area; Behavioral; Brain; Brain region; Clutterings; Code; Complex; Cues; Data; Dimensions; Discrimination; Disease; Electrophysiology (science); Environment; Etiology; Eye; Face; Feedback; Frequencies; Future; Goals; Human; Image; Impaired cognition; Industry; Joints; Knowledge; Macaca; Magnetic Resonance Imaging; Measures; Monkeys; Motor output; Neurons; Pathway interactions; Perception; Positioning Attribute; Process; Reliability of Results; Research; Retina; Retinal; Robot; Sensory; Signal Transduction; Stimulus; Structure; Testing; Uncertainty; Variant; Vision; Vision Disparity; Visual; Visual Perception; Visual system structure; Weight; Work; base; experimental study; falls; imaging approach; improved; insight; monocular; movie; neural circuit; neural correlate; neuroimaging; neurophysiology; orientation selectivity; public health relevance; receptive field; relating to nervous system; response; retinal imaging; retinotopic; sample fixation; stereoscopic; theories; three dimensional structure; two-dimensional; virtual reality; visual information

PROJECT SUMMARY/ABSTRACT How do we perceive the three-dimensional (3D) structure of the world when our eyes only sense two-dimensional (2D) projections like a movie on a screen? Reconstructing 3D scene information from 2D retinal images is a highly complex problem, made evident by the great difficulty robots have in turning visual inputs into appropriate 3D motor outputs to move physical chessmen on a cluttered board, even though they can beat the best human chess players. The goal of this proposal is to elucidate how hierarchical cortical circuits implement robust (i.e., accurate & precise) 3D visual perception. Towards this end, we will answer two fundamental questions about how the brain achieves the 2D-to-3D visual transformation using behavioral, electrophysiological, and neuro- imaging approaches. In Aim 1, we will answer the question of how the visual system represents the spatial pose (i.e., position & orientation) of objects in 3D space. Our hypothesis is that 3D scene information is reconstructed within the V1  V3A  CIP pathway. We will test this hypothesis by simultaneously recording 3D pose tuning curves from V3A and CIP neurons in macaque monkeys while the animals perform an eight-alternative 3D orientation discrimination task. This experiment will dissociate neural responses to 3D pose that reflect elementary receptive field structures (resulting in 3D orientation preferences that vary with position-in-depth, which we anticipate to find in V3A) from those that represent 3D object features (resulting in 3D orientation preferences that are invariant to position-in-depth, which we anticipate to find in CIP). Using these data, we will additionally test for functional correlates between neural activity in each area and perceptual sensitivity. Through application of Granger Causality Analysis to simultaneous local field potential recordings in V3A and CIP, we will further test for feedforward/feedback influences between the areas to evaluate their hierarchical structure. In Aim 2, we will answer the question of how binocular disparity cues (differences in where an object's image falls on each retina) and perspective cues (features resulting from 2D retinal projections of the 3D world) are integrated at the perceptual and neuronal levels to achieve robust 3D visual representations. Both cues provide valuable 3D scene information, and human perceptual studies show that their integration is dynamically reweighted depending on the viewing conditions (i.e., position-in-depth & orientation-in-depth) to achieve robust 3D percepts. Specifically, greater weight is assigned to the more reliable cue based on the viewing conditions; but, where and how this sophisticated integrative process is implemented in the brain is unknown. We anticipate that V3A and CIP will each show sensitivity to both cue types, but only CIP will dynamically reweight the cues to achieve robust 3D representations. This research is important for understanding ecologically relevant sensory processing and neural computations that are required for us to successfully interact with our 3D environment. Insights from this work will also extend beyond 3D vision by elucidating processes implemented by neural circuits to solve highly nonlinear optimization problems that turn ambiguous sensory signals into robust perceptions.

项目概要/摘要 当我们的眼睛只能感知二维时，我们如何感知世界的三维 (3D) 结构 （2D）像屏幕上的电影一样投影？从 2D 视网膜图像重建 3D 场景信息是 高度复杂的问题，从机器人将视觉输入转化为适当的信息的巨大困难就可以看出 3D 电机输出可在杂乱的棋盘上移动物理棋子，即使它们可以击败最好的人类 国际象棋棋手。该提案的目标是阐明分层皮层电路如何实现鲁棒性（即， 准确且精确）3D 视觉感知。为此，我们将回答两个基本问题： 大脑如何利用行为学、电生理学和神经学来实现 2D 到 3D 视觉转换 成像方法。在目标 1 中，我们将回答视觉系统如何表示空间姿态的问题 3D 空间中物体的位置（即位置和方向）。我们的假设是 3D 场景信息被重建 在V1→V3A→CIP途径内。我们将通过同时记录 3D 姿态调整来测试这个假设 猕猴执行八种替代 3D 操作时 V3A 和 CIP 神经元的曲线 方向辨别任务。该实验将分离对 3D 姿势的神经反应，这些反应反映了 基本感受野结构（导致 3D 方向偏好随位置深度而变化， 我们预计在 V3A 中找到的）来自那些代表 3D 对象特征的特征（导致 3D 方向） 位置深度不变的偏好，我们预计会在 CIP 中找到）。使用这些数据，我们将 另外测试每个区域的神经活动与知觉敏感性之间的功能相关性。通过 将格兰杰因果关系分析应用于 V3A 和 CIP 中同步局部场电位记录，我们将 进一步测试区域之间的前馈/反馈影响，以评估其层次结构。在 目标 2，我们将回答双目视差如何提示的问题（物体图像落在何处的差异） 每个视网膜上）和透视线索（3D 世界的 2D 视网膜投影产生的特征） 在感知和神经元层面进行整合，以实现强大的 3D 视觉表示。两种提示均提供 有价值的 3D 场景信息，人类感知研究表明它们的集成是动态的 根据观察条件（即深度位置和深度方向）重新加权以实现稳健 3D 感知。具体来说，根据观看条件，为更可靠的提示分配更大的权重； 但是，这种复杂的整合过程在大脑中的何处以及如何实施尚不清楚。我们预计 V3A 和 CIP 都会显示对两种提示类型的敏感性，但只有 CIP 会动态地重新加权提示 实现稳健的 3D 表示。这项研究对于理解生态相关的感官非常重要 我们成功地与 3D 环境交互所需的处理和神经计算。 通过阐明神经电路实现的过程，这项工作的见解也将超越 3D 视觉 解决高度非线性的优化问题，将模糊的感觉信号转化为稳健的感知。