Functional Anatomy of Perceptual and Attentional Systems in the Primate Brain

一个典型的场景包含许多不同的对象，由于视觉系统的处理能力有限，这些对象竞争神经表示。在神经层面上，多个物体之间的竞争表现为它们的视觉诱发反应的相互抑制。多个对象之间的竞争可能会受到自下而上的感官驱动机制(外源性注意)和自上而下的目标导向影响(如选择性内源性注意)的影响。虽然多个物体之间对表征的竞争最终在视觉皮质内解决，但自上而下的偏向信号的来源可能来自额叶和顶叶皮质区域的分布网络。 在过去一年里，我们完成了两项研究，并启动了另外两项研究。 在第一个已完成的研究中，我们探索了一个人如何关注另一个人的行为。使用心理物理学技术，并使用机器学习对人类伸手动作的视频进行广泛分析，我们表明，人类能够从身体微妙的准备动作中预测他人的动作。参与者被展示了从两个玩竞技或合作游戏的受试者中获取的达成任务的视频。视频在不同的时间点被剪辑，参与者被告知根据每个试验中早期的运动来预测伸展的目标。结果表明，人类可以从动作早期出现的微妙的准备动作中读出动作的目标。使用机器学习分析也得到了类似的结果。与竞争环境相比，准备活动在合作环境中表现出更多的变异性。然而，即使在竞争的背景下，参与者也有向对手隐瞒信息的动机，也有充分的线索背叛了他们的目标。这些结果表明，人类可能有一个身体运动的生物力学模型，帮助他们决定他人未来的行动方向。这些发现加深了我们对人类动作预测能力的理解，并为未来旨在理解动作预测的神经基础的神经科学和神经建模研究提供了信息。 在第二项已完成的研究中，我们探索了调节统计学习的神经底物，即人类和动物关注并提取其环境中的统计规律的过程。从人类参与者那里收集了行为和功能磁共振成像(FMRI)数据，同时他们观察刺激并将其归类为两种类别之一(例如，有生命的和无生命的)。这些刺激被分成不同的图案化或随机图像序列集合。对比随机和图案化图像序列的fMRI激活，发现体素在整个视觉皮质的随机序列中具有更大的激活，从早期的视觉区域延伸到腹侧颞叶。这些结果表明，对模式的视觉统计学习遵循预测编码模型，在该模型中，视觉皮质中对预测模式的反应存在衰减。 我们还启动了一个实验来研究视觉对象对抓取的加工。为了判断对象之间的相似性，我们注意到一组特征(几何、语义等)。当做出抓取物体的动作时，我们也注意到几何特征的一个子集。目前尚不清楚作为一般物体知识基础的相同机制是否也支持处理物体形状以进行掌握。在这个项目中，我们问：1)物体形状的哪些方面最能解释抓取运动？2)用于抓取的特征和用于一般物体相似性判断的特征之间有多少重叠？为了回答这些问题，我们记录了参与者对58个3D打印物体的抓取动作。我们使用这些动作来提取与抓取相关的特征。在另一项单独的实验中，我们要求其他参与者判断这些对象之间的相似性，然后使用这些数据来提取与一般对象相似性评级相关的特征。令人惊讶的是，与抓取相关的特征在很大程度上与用于一般相似性判断的特征不同，这表明提取这些特征的机制可能在大脑中有所不同。这些实验措施接下来将被用于功能磁共振实验，以识别处理物体特征以进行抓取的视觉大脑区域。 最后，第二个新启动的实验考察了枕颞叶和顶叶皮质在处理物体形状以计划抓握运动中的作用。我们设计了一种装置，允许参与者在核磁共振扫描仪内查看和掌握3D打印的物体。这些物体被放在一张放在身体上方的桌子上，参与者通过镜子观看物体。这些物品是由手柄部分(弯曲或直)和身体部分(圆形或矩形)组成的四个马克杯状物品之一。参与者被指示抓住手柄。我们聚焦于两个感兴趣的形状选择性区域，一个在枕叶外侧皮质(LOC)，另一个在顶下沟(下IPS)。对视觉反应进行模式分类分析，以区分具有相同身体但不同手柄的两个物体(手柄分类)或具有相同手柄但不同身体的两个物体(身体分类)。LoC对躯干和手柄的分类准确率相似，但下位IPS对手柄的模式分类显著高于躯干。这些结果表明，人类顶沟中的区域提取与物体正确互动相关的视觉信息。该项目的数据收集工作仍在进行中。

A typical scene contains many different objects that compete for neural representation due to the limited processing capacity of the visual system. At the neural level, competition among multiple objects is evidenced by the mutual suppression of their visually evoked responses. The competition among multiple objects can be biased by both bottom-up sensory-driven mechanisms (exogenous attention), such as stimulus salience, and top-down, goal-directed influences, such as selective, endogenous attention. Although the competition among multiple objects for representation is ultimately resolved within visual cortex, the source of top-down biasing signals likely derives from a distributed network of areas in frontal and parietal cortex. During the past year, we have completed two studies and initiated two others. In the first completed study, we explored how one attends to another's actions. Using psychophysics techniques and extensive analysis of videos of human reaching actions using machine learning, we showed that humans are able to predict the actions of others from subtle preparatory movements in the body. Participants were shown videos of a reaching task taken from pairs of subjects playing a competitive or cooperative game. Videos were cut at various time points and participants were told to predict the goal of the reach from early movements in each trial. Results showed that humans can read the goal of an action from subtle preparatory movements present early in the movement. Similar results were obtained using a machine learning analysis. The preparatory movements showed more variability in cooperative compared to competitive contexts. However, even in the competitive context, in which participants had the incentive to conceal information from their opponent, there were ample cues that betrayed their goals. These results suggest that humans may have a biomechanical model of body movements that help them determine the future course of actions of others. These findings deepen our understanding of human action prediction ability and inform future neuroscience and neural modeling research aimed at understanding the neural underpinnings of action prediction. In the second completed study, we explored the neural substrates mediating statistical learning, the process by which humans and animals attend to and extract the statistical regularities in their environment. Behavioral and functional magnetic resonance imaging (fMRI) data were collected from human participants, while they viewed and classified stimuli as belonging to one of two categories (e.g. animate and inanimate). The stimuli were grouped into different sets of either patterned or random image sequences. Contrasting fMRI activation during random compared to patterned runs of images, revealed voxels with greater activation for random runs throughout visual cortex, extending from early visual areas to the ventral temporal lobe. These results suggest that visual statistical learning of patterns follows predictive coding models in which there is an attenuation of the responses for predicted patterns in the visual cortex. We also initiated an experiment to investigate the processing of visual objects for grasp. To judge the similarity between objects, we attend to a collection of features (geometric, semantic, etc.). When making an action to grasp an object, we also attend to a subset of geometric features. It is not clear if the same mechanisms that underlie general object knowledge also support processing object shapes for grasp. In this project, we asked: 1) What aspects of an object shape can best explain grasping movements? and 2) How much overlap is there between features used for grasp and features used for general object similarity judgments? To answer these questions, we recorded participants' grasp movements towards 58 3D-printed objects. We used these movements to extract features relevant for grasp. In a separate experiment, we asked other participants to judge the similarity between these objects and then used these data to extract features relevant for general object similarity ratings. Surprisingly, the features relevant for grasp were largely distinct from those used for general similarity judgments, suggesting that the mechanisms for extracting these features may diverge in the brain. These experimental measures will be next used in an fMRI experiment to identify visual brain regions that process object features for grasp. Finally, a second newly initiated experiment examined the role of occipitotemporal and parietal cortices in processing the shape of objects to plan a grasp movement. We designed a set-up to allow participants to view and grasp 3D-printed objects inside an MRI scanner. The objects were put on a table positioned over the body, and participants viewed the object through a mirror. Objects were one of four mug-shaped items composed of a handle part (either curved or straight) and a body part (either round or rectangular. Participants were instructed to grasp the handle. We focused on two shape-selective regions of interest, one in the lateral occipital cortex (LOC), and one in the inferior intraparietal sulcus (inferior IPS). A pattern classification analysis was performed on the visual responses to discriminate either the two objects with the same body and different handles (handle classification) or the two objects with the same handle and different bodies (body classification). LOC showed similar classification accuracies for the body and handle, but inferior IPS showed significantly higher pattern classification for the handle than the body. These results suggest that regions in the human intraparietal sulcus extract visual information relevant for proper interaction with objects. The data collection for this project is still underway.