毕业生知识库

人类和非人灵长类动物实现感知、运动、推理等高级功能依赖于不同层次下的脑网络结构。脑网络是复杂的多尺度系统，在介观层面由空间上分离的神经元集群通过相互作用构成，而在宏观层面表现为脑区或亚区之间的相互作用。这些相互作用在不同层面下介导了信息的处理与传递，并形成了脑功能的神经基础。针对这两个层面的脑网络组织结构和动态活动特性，目前尚有一些基本问题未得到回答：一是宏观脑网络中，特别是在执行各种任务过程中的相互作用结构如何；二是在各层次的相互作用下脑网络如何根据不同的任务需求灵活地进行动态重组，以支持丰富的行为还远未明了。这些知识的缺乏是深入理解脑网络中的活动调控、信息处理以及网络机制的重要障碍。

针对第一个问题，本文第一项工作系统研究了宏观脑网络的相互作用结构。通过分析人脑在感知、运动、记忆等不同任务下的功能磁共振影像数据以及在选择注意任务下的头皮脑电数据，发现在这两种不同的时间尺度下，宏观功能网络不依赖高阶相互作用，即仅用二阶相互作用就能组织起丰富的功能网络。通过对功能磁共振影像数据的模拟和仿真分析，发现这种现象背后的机制是任务状态下宏观脑区或亚区之间的连接仍然为弱耦合。这项工作的研究结果首次揭示了在各种信息处理过程中，脑网络的动态活动是仅依靠二阶相互作用组织起来的。这一结果使脑网络分析的复杂度从2^N降低到N²。不仅为基于二阶相关性分析任务状态下的功能网络提供了合理性，也为简化人工智能系统的设计提供了借鉴。

针对第二个问题，本文第二项工作研究了跨模态选择注意任务下人脑宏观脑网络的重构机制；以及在第三项工作中，研究了视觉-运动映射任务下猕猴前额叶皮层介观网络的动态重组机制。在第二项工作中，基于人的头皮脑电数据，采用数据驱动的方法，发现α波段 (12-15 Hz) 的活动（格兰杰因果、功率谱）是支持跨模态视听觉选择注意的可靠的宏观网络特征。听觉注意诱发更强的α波段活动，特别是在枕顶区域。本研究的结果首次揭示了跨模态视听觉选择注意转换所对应的宏观网络重构模式，凸显了α波段神经同步活动在介导宏观网络重构从而实现跨模态视听觉注意转换的重要功能。

在第三项工作中，基于猕猴前额叶（46区）的局部场电位信号，研究了视觉-运动映射任务中介观网络的动态重组，发现前额叶网络在任务的不同阶段，由不同频段的神经同步活动主导。在运动执行阶段出现显著的β-γ (20-40 Hz) 同步活动增强，并且该频段下的效用性网路受结构性网络的制约，也受任务相关信息的调控。基于平均场模型的模拟结果显示，效用性网络的动态变化可以受特定频段输入电流的相位的影响。本研究揭示了前额叶皮层是通过利用不同频率下的神经振荡、以及效用网络的重构从而实现功能的灵活转换。

通过上述一系列研究，本文首先在宏观脑网络层面阐明了网络活动的基本组织结构，揭示了丰富的网络动态活动可以由较为简单的二阶相互作用介导；然后基于这种相互作用结构，研究了宏观及介观脑网络在具体行为任务下的动态重组规律，从而深化了对任务状态下多尺度脑网络的理解。

The high-level functions such as perception, motor and reason of humans and none-human primates depend on the structures of brain networks at different scales. The brain networks are multi-faceted, manifested on the mesoscopic level as the interaction among spatially separated neuronal clusters, and on the macroscopic level the interactive brain regions or subregions. These interactive structures not only mediate the information processing and relaying, but serve as the neural basis for brain functions. Regarding the brain networks on the macro- and meso- scopic level, two fundamental questions still remain unclear: one is the interactive structure on the macroscopic level, with the other being how the brain network flexibly reconfigures itself according to different task demands. The lack of the answers to these questions inhibits the deeper understanding towards the activity modulation, information processing, and network mechanism of the brain.

For the first question, this thesis firstly systematically studied the interaction structure of the macroscopic brain network. By analyzing the functional magnetic resonance imaging (fMRI) dataset collected from humans at different tasks, such as perception, movement and memory, as well as the electroencephalograph (EEG) data sampled during a selective attention task, it is revealed that for different temporal scales, rich functional networks can be organized regardless of high order interactions (HOIs). By dynamic modeling of the fMRI data, it is found that the weakly linear coupling strength among macroscopic regions or subregions is the underlying mechanism. This work demonstrates for the first time that pairwise interactions organize large-scale macroscopic functional networks at task state without involving HOIs. The complexity of network is reduced from 2^N to N², not only for a large number of network analysis based on pairwise interactions, but for the design of artificial intelligence system.

For the second question, the remaining part of the thesis studied the macroscopic network mechanism of human brain during a cross-modal selective attention task and the dynamic reconfiguration of macaque monkey’s prefrontal cortex at the mesoscopic level during a visual-motor mapping task. In the second work of this thesis, based on the EEG data, it is demonstrated by a data-driven method that the alpha-band (12-15 Hz) activity (Granger causality and power spectrum) is a robust network feature supporting the cross-modal audiovisual selective attention transition. Auditory attention induces stronger alpha activity, especially in the parietal-occipital areas. This work reveals for the first time the network mechanism regarding the cross-modal audiovisual selective attention, emphasizing the critical role played by alpha activity in mediating the attentional transition at the macroscopic network level.

In the third work, based on the local field potential collected in the macaque prefrontal lobe (area 46), the dynamic mesoscopic network in the visual-motor mapping task was revealed. Neural synchronization at different frequency bands dominates the network activity with the ongoing task. At the motor execution stage, there is a significant increase of beta-gamma (20-40 Hz) activity. The effective connectivity at this frequency band was found to be constrained by the structural network, and modulated by the task-related information. The simulation results based on mean field model further showed that the dynamic effective network can be affected by the phase of input current at specific frequency band. This study illustrates that the flexible role played by prefrontal cortex is supported by neural oscillations at different frequency bands and the dynamic reconfiguration of effective network.

Through the above analysis, this thesis firstly clarified the basic organization structure of network activities on the macroscopic level of brain network, and revealed that rich activity patterns can be mediated by relatively simple, merely second-order interaction. Then, based on this interaction structure, the dynamic reconfiguration rules of macroscopic and mesoscopic brain networks at specific behavioral tasks are studied, thus deepening the understanding of the activities of multi-scale brain networks at task states.