CRCNS: Collaborative Research: Dynamic Models of Human Auditory Perceptual Switching Informed by Large-Scale ECoG Recordings

自然环境中的声音是来自许多不同来源的复杂混合物。该项目旨在了解人类如何将混合的声音组织成有意义的物体。对听觉对象的感知不是来自大脑的任何特定部分，而是来自许多大脑区域的协调活动；此外，声音与听觉对象的绑定可能会非常迅速地转换。因此，研究听觉对象是如何形成的，以及快速转换是如何发生的，需要以非常高的速度分析人类大脑许多区域的大脑活动记录。本项目旨在开发新的理论方法来整合和分析来自大规模电极阵列的复杂动态大脑记录数据集。建模方法将为理解正常和临床受损大脑的人类听觉提供见解。在过去的三十年中，听觉感知的神经相关特征在听觉皮层的研究取得了重大进展。然而，这些神经关联可能并不局限于听觉皮层，或大脑的任何特定部分。为了理解听觉感知表征和感知转换的神经机制，本项目结合了实验设计和理论两方面的进展。在涉及快速感知转换的双稳态听觉任务中，当受试者自我报告他们的感知时，将收集大规模的皮质电图（ECoG）记录。接下来，将使用一种数据驱动的神经科学新方法——动态模式分解（DMD），从这些大型时间序列数据集中提取任务期间皮层激活的时空模式。然后，DMD提取的特征将用于构建数据驱动的低维动态模型，该模型捕捉多个皮层区域的时间演变，编码听觉刺激和感知状态。

Sounds in natural environments are complex mixtures from many different sources. This project seeks to understand how humans organize mixtures of sounds into meaningful objects. Perceptions of auditory objects arise not from any particular part of the brain, but rather from coordinated activity across many brain regions; further, binding of sounds to auditory objects may switch very rapidly. Therefore, the study of how auditory objects are formed and how rapid switching occurs requires analyzing recordings of brain activity in humans across many brain areas and at very high speed. This project aims to develop new theoretical methods for integrating and analyzing complex dynamic data sets of brain recordings from large-scale electrode arrays. The modeling approach will provide insight in the understanding of human auditory perception in both normal and clinically impaired minds. Significant advances have been made in the past three decades characterizing neural correlates of auditory perceptions localized to the auditory cortex. Nevertheless, these neural correlates are likely not restricted to the auditory cortex, or to any particular part of the brain. To understand the neural mechanisms of auditory perceptual representation and perceptual switching, the current project combines advances in both experimental design and theory. Large-scale electrocorticography (ECoG) recordings will be collected from human subjects as they self-report their perceptions during a bistable auditory task involving rapid perceptual switching. Next, spatial-temporal patterns of cortical activation during the task will be extracted from these large time-series datasets using a data-driven method novel to neuroscience known as dynamic mode decomposition (DMD). Features extracted by DMD will then be used to build data-driven, low-dimensional dynamic models that capture the temporal evolution of multiple cortical areas, encoding both the auditory stimulus and the perceptual state.