Neuromodeling, Functional Brain Imaging And Language

该项目的重点是了解大脑如何构建相互作用区域的网络（即神经网络）来执行认知任务，特别是与语言相关的任务，以及这些网络在大脑疾病中如何改变。这些问题是通过结合计算神经科学技术和功能性神经成像数据来解决的，这些数据是通过PET或fMRI获得的（参见神经网络13:829-840,2000）。网络分析方法使我们能够评估不同任务之间以及正常人群和患者之间的大脑操作差异。这项研究将使我们能够确定哪些网络功能失调，以及神经可塑性在使代偿行为发生方面所起的作用。我们已经开始用PET测量区域脑血流量（rCBF）（一种神经活动指数）来描绘正常受试者自发叙述性言语产生的功能网络。我们确定了左半球（LH）大脑区域的功能连通性（以不同脑区rCBF之间的相关性来评估）。我们的研究结果表明，在正常的受试者中，在语言产生过程中，LH - perisylvian区域之间相互作用强烈，但在需要类似于说话的肌肉运动和发声的任务中却没有。在口吃患者中，许多这些强大的功能联系似乎缺失，这表明在口吃期间LH语言产生网络是异常的。我们还使用功能磁共振成像检查了大脑功能连接。研究了左额下回两个区域在单词、假单词、辅音串和假字体四种视觉刺激的正字法加工过程中的功能连通性。我们发现，只有在处理语义丰富的刺激（如单词）时，上颌前叶前部前部与枕叶皮层左后部区域之间才有很强的功能连通性。相反，当刺激具有语音内容（即单词、假词和字母串）时，前叶前叶前部后部与左枕后区和颞叶区有很强的功能连接。这些结果支持了腹侧LIFG更专注于语义处理，而后侧/背侧LIFG更专注于语音处理的观点（Neuron 30: 609-617, 2001）。对于语言功能，LIFG的一个重要部分是Broca区，它被定义为Brodmann设计的系统中的细胞结构区44和45。我们使用了与细胞结构定义的Brodmann区域（BA） 44和45相对应的概率数据集，这使我们能够说出Talairach地图集立体定向空间中的任何给定体素，该体素在BA44或BA45中的概率是多少。我们将这一概率图谱应用于在语言生成任务（叙述性语言的生成）中获得的PET数据，这些数据来自父母是聋人的成年人，他们的语言和美国手语（ASL）都很流利。叙事性的制作是在单独的PET扫描中使用语音和手势进行的。我们发现言语和手语在布洛德曼区域的中心部位都有类似的激活，这表明尽管言语和手语的表达方式不同，但布洛德曼区域的神经基质是相同的。为了了解功能性神经成像研究中观察到的结果与潜在的神经动力学之间的关系，我们之前构建了一个大规模的神经元动力学计算机模型，该模型执行视觉对象匹配任务，类似于为PET研究设计的任务（《大脑皮层》8:31 - 20,1998）。该模型由与大脑皮层神经元集合相对应的元素组成，并包含基于猴子在执行类似任务时的电生理记录所识别的类型的不同元素。它包括一个涉及枕颞视觉通路和额叶回路的“活跃”记忆网络，能够执行匹配样本任务，如果第二个刺激与第一个刺激匹配，就会做出反应。PET研究是通过向代表外侧膝状核的模型区域呈现成对刺激来模拟的。模拟PET数据从模型中计算出来，因为它通过整合不同区域内的突触活动来执行任务。模拟的PET数据类似于在实际的PET延迟匹配样本视觉任务中发现的数据，以及每个大脑区域正确的神经元动态。在过去的一年里，我们研究了神经抑制如何在PET/fMRI信号中表现出来。我们发现，神经元抑制可以导致PET/fMRI测量的活动增加或减少，这取决于存在的兴奋性活动的数量（Brain Res. Bull. 54: 267-273, 2001）。我们还研究了门控机制如何调节短期记忆的参与和保留，并对我们的额叶工作记忆回路进行了改进，以更好地反映多巴胺等神经调节剂的已知特性（神经网络13:941-952,2000）。我们也开始扩展模型，以便它也可以模拟听觉处理（国际神经网络联合会议论文集，出版中）。我们构建了皮层听觉模式识别（what）通路的大规模神经模型，该神经模型包括4个区域：初级听觉皮层（Ai）、次级听觉皮层（Aii）、颞上回（STG）和前额叶皮层（PFC）。模式前3个阶段存在不同尺度的时间整合，其中Ai的时间窗口最小，STG的时间窗口最长。与由稳态音调组成的控制任务相比，瞬态音调模式导致所有区域的模拟rCBF增加，其中STG和PFC区域的增加最多，这表明刺激的复杂性与高级听觉处理区域的更大激活相关。这项建模工作将与模拟研究中使用的相同任务的PET和fMRI实验相结合。我们还将一种模拟经颅磁刺激（TMS）的方法纳入了大尺度模型。在经颅磁刺激过程中，一个强大的、不断变化的磁场作用于头皮上的某个区域，诱发颅内电流，从而改变该区域的神经元功能。经颅磁刺激对受刺激的神经组织具有兴奋和抑制作用，尽管目前对经颅磁刺激改变神经元功能的确切神经生物学机制知之甚少。经颅磁刺激已与PET结合使用来检查人类大脑皮层的区域间连通性。在本实验中，我们将经颅磁刺激应用于视觉处理通路的大尺度神经模型，以研究其对与刺激区域相连的rCBF的影响。在实验研究中，观察到经颅磁刺激后rCBF的增加和减少。在该模型中，当颅磁刺激主要发挥兴奋作用时，颅磁刺激强度增加导致rCBF增加，而当颅磁刺激主要发挥抑制作用时，颅磁刺激后rCBF减少。我们还发现直接和间接连接到刺激部位的区域都受到TMS的影响（NeuroImage, press）。

This project focuses on understanding how the brain constructs networks of interacting regions (i.e., neural networks) to perform cognitive tasks, especially those associated with language, and how these networks are altered in brain disorders. These issues are addressed by combining computational neuroscience techniques with functional neuroimaging data, obtained using PET or fMRI (reviewed in Neural Networks 13: 829-840, 2000). The network analysis methods allow us to evaluate how brain operations differ between tasks, and between normal and patient populations. This research will allow us to ascertain which networks are dysfunctional, and the role neural plasticity plays in enabling compensatory behavior to occur. We have begun delineating the functional networks involved in the production of spontaneous narrative speech in normal subjects using PET measurements of regional cerebral blood flow (rCBF), an index of neural activity. We determined the functional connectivity (evaluated as the correlation between rCBF in different brain areas) of left hemisphere (LH) brain regions. Our results demonstrate that in normal subjects LH perisylvian regions interact strongly with one another during language production, but not during a task requiring similar muscle movements and vocalizations as speech. In patients who stutter, many of these strong functional linkages seem to be absent, suggesting that LH language-production networks are abnormal during stuttering. We also examined brain functional connectivity using fMRI. We investigated the functional connectivity of two areas of the left inferior frontal gyrus (LIFG) during orthographic processing of four kinds of visual stimuli - words, pseudowords, consonant strings and false fonts. We found strong functional connectivity between ventral LIFG and left posterior areas in occipital and temporal cortex only during the processing of semantically rich stimuli (i.e., words). Conversely, the posterior/dorsal part of LIFG had strong functional connections with left posterior occipital and temporal areas when the stimuli had phonological content (i.e., words, pseudowords and letter strings). These results support the notion that ventral LIFG is more specialized for semantic processing and that posterior/dorsal LIFG is more engaged in phonological processing (Neuron 30: 609-617, 2001). For language function, one important part of the LIFG is Broca's area, which is defined as the cytoarchitectonic areas 44 and 45 in the system devised by Brodmann. We used a probabilistic data set corresponding to cytoarchitectonically-defined Brodmann areas (BA) 44 and 45, which enables us to say for any given voxel in the stereotactic space of the Talairach atlas, what the probability is that this voxel is in BA44 or BA45. We applied this probabilistic atlas to PET data acquired during language production tasks (generation of narrative speech) from adults whose parents were deaf and who were fluent in both speech and American Sign Language (ASL). Narrative production was performed in separate PET scans using speech and sign. We found similar activations of the central parts of both Brodmann areas by both speech and sign, suggesting that in spite of the different modalities by which speech and sign are expressed, the same neural substrates in Broca's region are engaged. In order to understand the relationship between what is observed in functional neuroimaging studies and the underlying neural dynamics, we had previously constructed a large-scale computer model of neuronal dynamics that performs a visual object-matching task similar to those designed for PET studies (Cerebral Cortex 8:310-20, 1998). The model is composed of elements that correspond to neuronal assemblies in cerebral cortex, and contain different elements that are based on types identified by electrophysiological recordings from monkeys as they perform similar tasks. It includes an "active" memory network involving the occipitotemporal visual pathway and a frontal circuit, and is capable of performing a match-to-sample task in which a response is made if the second stimulus matches the first. A PET study is simulated by presenting pairs of stimuli to an area of the model that represents the lateral geniculate nucleus. Simulated PET data are computed from the model as it performs the tasks by integrating synaptic activity within the different areas. Simulated PET data similar to that found in actual PET delayed match to sample visual tasks were obtained, as were the correct neuronal dynamics in each brain region. In the past year, we investigated how neural inhibition is manifested in PET/fMRI signals. We found that neuronal inhibition can cause measured PET/fMRI activity to either increase or decrease, depending on the amount of excitatory activity present (Brain Res. Bull. 54: 267-273, 2001). We also investigated the way gating mechanisms can regulate engagement and retention of short-term memory, and refinements were made in our frontal working memory circuit to better reflect known properties of neuromodulators such as dopamine (Neural Networks 13: 941-952,2000). We have also begun an expansion the model so that it can also simulate auditory processing (Proceedings of the International Joint Conference on Neural Networks, in press). We constructed a large-scale neural model of the auditory pattern recognition ('what') pathway in cortex consisting of 4 regions: Ai (primary auditory cortex), Aii (secondary auditory cortex), STG (superior temporal gyrus) and PFC (prefrontal cortex). There are different scales of temporal integration in the first 3 stages of the model, with Ai having the smallest temporal window and the STG the longest. Compared to a control task consisting of steady-state tones, transient tonal patterns resulted in increased simulated rCBF in all regions, with the highest increases occurring in the STG and PFC regions, suggesting that the complexity of the stimulus is correlated with greater activation in higher auditory processing areas. This modeling effort will be combined with PET and fMRI experiments of the same kind of tasks used in the simulation studies. We also have incorporated into the large-scale model a way to simulate transcranial magnetic stimulation (TMS). During TMS, a strong, changing magnetic field applied to a region on the scalp induces intracranial electrical currents that can alter regional neuronal function. TMS exerts both excitatory and inhibitory effects on stimulated neural tissue, although little is known about the exact neurobiological mechanisms by which TMS alters neuronal function. TMS has been used in conjunction with PET to examine interregional connectivity of human cerebral cortex. In the present simulations, TMS was applied to the large-scale neural model of the visual processing pathway to investigate its effects on rCBF in regions connected to the stimulated area. In experimental studies, both increases and decreases in rCBF following TMS have been observed. In the model, increasing TMS intensity caused an increase in rCBF when TMS exerted a predominantly excitatory effect, whereas decreased rCBF following TMS occurred if TMS exerted a predominantly inhibitory effect. We also found that regions both directly and indirectly connected to the stimulating site were affected by TMS (NeuroImage, in press).