Object, face, body and scene representations in the human brain

这项研究的目的是了解我们如何看到我们所看到的：大脑如何分析落在眼睛视网膜上的光线，以揭示一个充满人，地方和事物的世界？ 正在进行的研究的主要领域之一集中在理解对象的多维表示（NCT 00001360）。 现实世界的场景是非常复杂和异构的，但我们能够对它们进行分类，并毫不费力地识别这些场景中的对象和人。虽然先前的研究已经确定了似乎专门用于处理面部，物体和场景的大脑区域，但仍不清楚这些不同区域的确切作用以及它们包含的信息。 理解大脑中视觉感知的主要挑战之一是我们所经历的各种不同的物体和场景。尽管如此广泛，研究往往使用少量的手选择的对象或场景类别，但它变得不清楚如何代表现实世界的处理随后的研究结果。为了克服这一挑战，我们开发了一个大型数据库（THINGS），其中包含1,854种不同类型的对象，这些对象从美国英语中具体的可图示和可命名的名词中系统地采样。THINGS数据库提供了对象概念和对象图像的丰富资源，并为心理学，神经科学和计算机科学领域的系统和大规模自然主义研究提供了工具。 我们在一个大规模的行为实验中使用了这个数据库，这个实验使用了在线众包，在5,000多名参与者中抽样了146万次试验。使用任务的计算模型，我们能够识别出我们对物体的内部心理表征的49个核心维度，为这个物体表征空间提供了全面和细粒度的表征。我们最近扩展了这项工作，收集了额外的300万次数据试验，将维度数量扩展到66个。 除了行为数据外，我们还收集了大量的功能性MRI和MEG数据，同时人们从THINGS数据库中查看图像。我们基于先前收集的行为反应（如上所述）开发的模型为现在研究对象的潜在神经表征提供了一个关键框架。对fMRI数据的分析表明，个体维度在视觉皮层（包括早期视觉皮层和高级皮层）以及顶叶和额叶区域广泛存在。此外，这些行为相关的尺寸是上级的对象类别信息在预测大脑皮层的反应。MEG数据的分析表明，各个维度迅速表示，开始约80-100毫秒，不同的维度显示不同的时间动态，可能反映的视觉与概念特征的对象的表示。 为了研究对物体的记忆，并确定是什么让一些图像比其他图像更容易记住，我们收集了一个额外的数据集，其中包含了超过一百万个识别记忆判断。使用这些数据，我们建立了一个预测图像记忆性的对象特征模型。我们发现，概念特征比视觉特征对我们记忆的影响更大。 总的来说，这些多维对象表征的研究为支持我们理解对象的认知过程提供了重要的见解。 阐明大脑如何使我们能够识别物体，场景，面孔和身体，为我们对周围世界的内部表征的本质提供了重要的见解。了解这些表征对于识别和表征许多精神健康和神经系统疾病的潜在缺陷至关重要。

The goal of this research is to understand how we see what we see: how does the brain analyze the light falling on the retina of the eye to reveal a world full of people, places, and things? One of the major areas of ongoing research focuses on understanding the multidimensional representations of objects (NCT00001360). Real-world scenes are incredibly complex and heterogeneous, yet we are able to categorize them and identify objects and people within those scenes effortlessly. While prior studies have identified brain regions that appear to be specialized for processing faces, object and scenes, it remains unclear what the precise roles of these different regions are and what information they contain. One of the major challenges in understanding visual perception in the brain is the wide range of different objects and scenes that we experience. Despite this breadth, studies often use a small number of hand selected object or scene categories, but it becomes unclear how representative of real-world processing the subsequent findings are. To overcome this challenge, we developed a large-scale database (THINGS) of 1,854 diverse types of object sampled systematically from concrete picturable and nameable nouns in the American English language. The THINGS database provides a rich resource of object concepts and object images and offers a tool for both systematic and large-scale naturalistic research in the fields of psychology, neuroscience, and computer science. We used this database in a large-scale behavioral experiment using online crowdsourcing, sampling 1.46 million trials in more than 5,000 participants. Using a computational model of the task, we were able to identify 49 core dimensions of our internal mental representations of objects, providing a comprehensive and fine-grained characterization of this object representational space. We have recently extended this work, collecting an additional 3 million trials of data, which expanded the number of dimensions to 66. In addition to the behavioral data we also collected extensive functional MRI and MEG data while people viewed images from the THINGS database. The model we developed based on the previously collected behavioral responses (described above) provides a critical framework for now investigating the underlying neural representations of objects. Analysis of the fMRI data reveals that the individual dimensions are represented broadly across visual cortex, including both early visual and high-level cortex, as well in parietal and frontal regions. Further, these behaviorally relevant dimensions were superior to object category information at predicting cortical brain responses. Analysis of the MEG data reveals that the individual dimensions are rapidly represented, starting around 80-100 ms, with different dimensions showing different temporal dynamics that may reflect the representation of visual versus conceptual features of the objects. To investigate memory for objects and determine what makes some images more memorable than others, we collected an additional dataset of more than one million recognition memory judgments. Using these data, we built a model of object features that is predictive of image memorability. We found that conceptual features exert a stronger influence than visual features on what we remember. Collectively, these studies of multidimensional object representations provide important insights into the cognitive processes supporting our understanding of objects. Elucidating how the brain enables us to recognize objects, scenes, faces and bodies provides important insights into the nature of our internal representations of the world around us. Understanding these representations is vital to identify and characterize the underlying deficits in many mental health and neurological disorders.