fMRI Reverse Correlation as a Novel Method for Revealing Computations Underlying Perceptual Grouping

fMRI 逆相关作为揭示感知分组基础计算的新方法

• 批准号: 2122866
• 负责人: Adam Greenberg
• 金额: $ 69.51万
• 依托单位: Medical College of Wisconsin
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2122866&HistoricalAwards=false
• 关键词: fMRI; Reverse; Correlation; Novel; Method

How do we segregate the distinct objects in a complex visual scene? Since most real-world objects are opaque, and therefore partly occlude one another, the eye receives a patchwork of overlapping surfaces, and the brain is then left with the task of determining how to perceptually group these various surface patches into unified objects. Although we do this effortlessly every day, we still do not understand the underlying neural computations that accomplish this scene analysis. Perceptual grouping cues (e.g., two surface fragments moving together or possessing the same texture) provide important clues that can be used to organize the visual scene into complete objects, but the specific computations performed, and the brain regions involved, are largely unknown. This project employs brain imaging to quantify the relative strength (and “pecking order”) of the many possible perceptual grouping cues used in constructing perceived objects from their component structures. This has been achieved by the development of a novel method for visual stimulation and analysis using noise-based image classification (i.e., reverse correlation) during functional brain imaging. This method has been used extensively in behavioral laboratory testing, but until recently, has not been practical for application to brain imaging because it typically requires a very large number of trials. However, by optimizing this technique to achieve reverse correlation during brain imaging, it is possible to uncover the brain regions driving the perception of objects in our environment. A more complete understanding of the brain mechanisms underlying perceptual grouping will lead to optimized designs for visual displays in our environment including street signs, occupational safety warnings, medical equipment instructions, and virtually all dynamic displays of visual information, as well as better artificial intelligence and robotic visual scene analysis, crucial for new technologies such as driverless cars. Neuroscientific studies of object perception have previously focused primarily on the specificity of object representations in the brain. In contrast, the new approach in this research is to study the psychological and neural underpinnings of the formation of these object percepts. A key innovation is the development of a novel quantitative metric to reliably and quantitatively measure perceptual grouping that is flexible enough to be used both behaviorally and during functional magnetic resonance brain imaging (fMRI). Using this approach, it is possible to determine the critical grouping cues for object perception and detail their dominance relations in careful behavioral testing, and then to adapt the reverse correlation method to be used with brain imaging data and optimize the algorithm to reduce the number of trials (and total brain scan time) required. Finally, this new technique, comparing internal templates of neural structures to behavioral templates, can be utilized to specify the network computations driving brain-behavior relations during perceptual grouping. The results of this research will advance our understanding of visual cognition, and resolve where in the brain, and specifically at which level of the visual processing cortical hierarchy, the visual grouping cues are operating. This research will reveal computational algorithms used by the human brain for perceptual grouping and scene segregation that can also be utilized to enhance artificial intelligence (AI) visual scene analysis.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

我们如何在复杂的视觉场景中分离不同的对象？由于大多数现实世界的物体都是不透明的，因此部分地相互遮挡，眼睛接收到的是重叠表面的拼凑而成，然后大脑的任务是确定如何在感知上将这些不同的表面拼凑成统一的物体。尽管我们每天都毫不费力地这样做，但我们仍然不了解完成此场景分析的底层神经计算。感知分组线索（例如，两个表面碎片一起移动或拥有相同的纹理）提供了重要的线索，可用于将视觉场景组织成完整的对象，但所执行的具体计算以及所涉及的大脑区域在很大程度上是未知的。该项目采用大脑成像来量化许多可能的感知分组线索的相对强度（和“啄食顺序”），这些感知分组线索用于从其组成结构构建感知对象。这是通过开发一种在功能性脑成像过程中使用基于噪声的图像分类（即逆相关）进行视觉刺激和分析的新方法来实现的。这种方法已广泛用于行为实验室测试，但直到最近，还没有实际应用于大脑成像，因为它通常需要大量的试验。然而，通过优化这项技术以在大脑成像过程中实现反向相关，有可能揭示驱动我们环境中物体感知的大脑区域。对感知分组背后的大脑机制的更全面理解将有助于优化我们环境中的视觉显示设计，包括街道标志、职业安全警告、医疗设备说明和几乎所有视觉信息的动态显示，以及更好的人工智能和机器人视觉场景分析，这对于无人驾驶汽车等新技术至关重要。物体感知的神经科学研究以前主要集中在大脑中物体表征的特异性。相比之下，这项研究的新方法是研究这些物体感知形成的心理和神经基础。一项关键的创新是开发了一种新颖的定量指标，可以可靠且定量地测量感知分组，该指标足够灵活，可以在行为上和功能性磁共振脑成像 (fMRI) 过程中使用。使用这种方法，可以确定物体感知的关键分组线索，并在仔细的行为测试中详细说明它们的主导关系，然后调整逆相关方法与大脑成像数据一起使用，并优化算法以减少所需的试验次数（和总大脑扫描时间）。最后，这种新技术将神经结构的内部模板与行为模板进行比较，可用于指定在感知分组期间驱动大脑行为关系的网络计算。这项研究的结果将增进我们对视觉认知的理解，并解决视觉分组线索在大脑的哪个位置，特别是在视觉处理皮质层次结构的哪个级别上起作用的问题。这项研究将揭示人脑用于感知分组和场景分离的计算算法，这些算法也可用于增强人工智能 (AI) 视觉场景分析。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。