Hemispheric and topographic neural organization of high-level visual representations

人们以惊人的速度和准确度识别人脸、单词和其他常见物体。 这个项目解决了面部，单词和物体的视觉知识通过学习在大脑中组织起来的方式。 先前的研究表明，右大脑半球专门用于表征面部，而左大脑半球对单词有更强的表征，两个半球对物体的表征大致平衡。 这些大脑半球专门化的差异是一个程度的问题，并且在个体之间存在差异。对这些差异的研究需要对识别能力进行行为测试，结合大脑结构和大脑激活模式的神经成像，以了解视觉识别系统是如何在两个大脑半球内部和之间组织起来的，这种组织是如何随着视觉体验出现的，以及它如何以及为什么在个体之间变化。与人类行为和神经成像研究平行，该项目的另一个方面是开发视觉识别系统的计算模拟，使用人工神经网络学习识别人脸，单词和物体以及人。这个模拟模型是为了模仿大脑的特性而设计的。该模型的变体可以重现行为，甚至可以捕捉不同个体大脑组织的潜在差异。该项目推进了我们对获得面部和单词识别的神经基础和潜在大脑组织的理解。这项研究对数百万需要矫正阅读障碍等发育障碍的美国人具有深远的治疗意义，他们需要帮助来克服字母和单词识别的困难，以及中风或神经外科手术后阅读和语言能力的恢复。视觉识别是由两个大脑半球的大脑区域网络支持的，该网络从幼儿时期的一些初始结构开始，然后通过经验发展到对不同类型的刺激具有分级专业化。 为了更好地理解识别是如何发生的，这项大规模的研究调查了大脑结构（例如，白色物质连通性）、大脑功能（例如，神经激活的选择性）和行为（例如，其中视觉刺激例如脸、词和物体呈现给一个半球）。这种多管齐下的方法使我们能够测试这样一种预测，即半球内连接性更强和/或半球间连接性更弱的参与者将对左半球的面孔和右半球的单词表现出更大的偏侧化（但对物体没有差异）。第二项研究测试了大脑语言侧化的差异性（在右撇子和左撇子之间）解释了单词侧化的个体差异，而这反过来又影响了面部侧化。 最后一项研究考察了当个体接触和学习新的视觉对象类别时大脑活动的细粒度变化，使我们能够测试与已知类别（面孔和单词）相比，在表示新信息时的竞争预测。 在这些研究的同时，本研究还包括通过训练一个空间受限的多层人工神经网络来识别大量的人脸、单词和物体，从而开发一个视觉识别系统的计算模型。该模型的性能将根据人类研究的数据进行评估，以确定该模型是否可以重新创建和解释所研究系统的各个方面，包括不同个体的不同模式。 在确定高层次视觉信息组织的基本原则时，这项工作对知识获取和学习具有重要意义，无论是在典型的个体生命周期中，该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查进行评估来支持的搜索.

People recognize faces, words, and other common objects with remarkable speed and accuracy. This project addresses the way that visual knowledge of faces, words, and objects becomes organized in the brain through learning. Prior research has shown that the right cerebral hemisphere is specialized for the representation of faces, whereas the left cerebral hemisphere has a stronger representation for words, with the representation of objects approximately balanced across both hemispheres. These differences in hemispheric specialization are a matter of degree and vary across individuals. Investigation of these differences requires behavioral testing of recognition abilities, combined with neuroimaging of brain structure and brain activation patterns, to understand how the visual recognition system is organized both within and between the two cerebral hemispheres, how this organization emerges with visual experience, and how and why it varies across individuals. In parallel with human behavioral and neuroimaging studies, another facet of this project is the development of a computational simulation of the visual recognition system, using an artificial neural network that learns to recognize faces, words and objects as well as people do. This simulation model is designed to mimic the properties of the brain. Variants of the model can reproduce the behavior and even capture the underlying differences in brain organization of different individuals. This project advances our understanding of the neural basis and underlying brain organization for acquisition of face and word recognition. This research has profound therapeutic implications for millions of Americans who require remediation of developmental disorders like dyslexia, who need assistance to overcome difficulties in letter and word recognition, and remediation for recovery of reading and language abilities after stroke or neurosurgery. Visual recognition is supported by a network of brain regions in both cerebral hemispheres, that starts with some initial structure in early childhood and then develops through experience to have graded specialization for different types of stimuli. To understand better how recognition occurs, this large-scale study investigates brain structure (e.g., white matter connectivity), brain function (e.g., selectivity of neural activation) and behavior (e.g., with visual stimuli such as faces, words, and objects presented to one hemisphere) over a large group of subjects. This multi-pronged approach allows us to test the prediction that participants with greater within-hemisphere and/or weaker between-hemisphere connectivity will show greater lateralization for faces to the left hemisphere and words to the right hemisphere (but no difference for objects). A second study tests predictions that variability in language lateralization of the brain (across right- and left-handed individuals) explains individual differences in word lateralization, and this, in turn, influences face lateralization. A final study examines fine-grained changes in brain activity as individuals are exposed to, and learn, novel visual object categories, allowing us to test predictions about competition in representing new information compared with known categories (faces and words). In parallel with these studies, this research includes the development of a computational model of the visual recognition system by training a spatially constrained multi-layer artificial neural network to recognize a large number of faces, words, and objects. The model’s performance is evaluated against data of the human studies to determine whether this model can re-create and explain aspects of the system under investigation, including the distinct patterns of different individuals. In identifying the principles underlying the organization of high-level visual information, the work has important implications for knowledge acquisition and learning, both in typical individuals across the lifespan, and in those facing difficulties due to developmental or neuropsychological disorders.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.