大脑中的Gamma振荡(25-100Hz)是一种常见于许多脑区的同步神经活动, 它被认为是大脑网络的一种功能性特征，与诸如注意、记忆、学习等正常认知功能紧密相关，在大脑信息处理（如视觉皮层）中也起着重要作用，并且异常的Gamma振荡常被发现与精神疾病有关。所以对Gamma振荡的功能和环路机制的理解，是解读大脑信息整合机制中的一个前沿问题。尽管Gamma被认为与视觉认知功能有关，但它所表征的视觉信息和产生的神经环路机制仍然不清楚或存在争议，因此本文不针对Gamma振荡的认知功能，而是针对视皮层中Gamma振荡的视觉编码规律以及视觉皮层的不同环路连接模式如何产生和调制Gamma振荡的问题，以麻醉猫为实验对象，结合电生理记录（在视觉脑区（外侧膝状体（LGN）和皮层18区（V1）））及数学建模（大尺度神经元网络）等手段，开展以下三方面的研究分别揭示视觉系统中Gamma振荡的视觉反应特性和环路机制。

研究一：针对前人发现的一个令人费解的现象，起初前人研究表明在视觉皮层中高空间频率视觉刺激可以诱发很强的Gamma振荡，然而近年来有其他研究者发现了不同的结果，实验观察到低空间频率视觉刺激也能诱发很强的Gamma振荡。为了解释这个分歧，研究一对不同空间频率诱发Gamma振荡的情况，以及信息如何通过Gamma进行神经活动编码进行探索，首次发现在V1中存在三种不同的窄带Gamma振荡，它们分别处理不同的空间频率（SF）信号，低频Gamma（LG，25-40hz）倾向于高空间频率和刺激边界，中频Gamma（MG，40-65 Hz）和高频Gamma（HG，65-90 Hz）倾向于中低空间频率和刺激表面。

研究二：基于研究一中发现的三个不同Gamma振荡，进一步探究他们各自的神经环路机制。以往研究认为，Gamma振荡在视觉系统的起源有两种可能，一种是Gamma产生于皮层下（包括视网膜和外膝体）；另外一种是Gamma并非皮下机制，而是产生于初级视觉皮层。因此在实验上，首先通过对于V1的分层分析，发现三个Gamma振荡不同的空间分布，LG在表层最强，HG在输入层最强，MG介于两者之间。HG在输入层最强的现象暗示着它可能有皮下的起源机制，之后通过同时记录初级视觉皮层（V1）和外侧膝状体（LGN）神经反应，发现高频Gamma振荡（HG）起源于皮下，而中频（MG）和低频（LG）的Gamma振荡起源于皮层。

研究三：进一步探索了起源于皮层的中频（MG）和低频（LG）的Gamma振荡的神经环路机制。在视觉皮层中有丰富的环路类型，而这些神经环路如何影响Gamma振荡的产生和调制还不完全清楚，为了回答这个问题，我们建立了一个大规模神经网络模型来仿真初级视觉皮层（V1）。这个数学模型在V1中具有生物学上合理的连接模式：局部环路连接（RC）、前馈连接（FF）、反馈连接（FB）和长程水平连接（HC）。研究表明中频Gamma振荡由局部连接环路产生，而低频的Gamma振荡由长程水平连接环路产生，它们同时会受到反馈连接的调制。

综上所述，本论文对Gamma的研究整合了前人多个分歧的实验结果（空间频率的反应特性与不同Gamma振荡的神经环路起源），这些发现表明，Gamma振荡反映了神经环路的神经动力学，这些神经环路在视觉系统中处理不同的空间频率信息，同时也是不同空间尺度上神经连接的特征。这些结果不仅帮助我们深入理解生物脑信息编码和整合的计算原理，对理解人脑的环路结构和认知功能具有重要的指导意义，还将为类脑计算的优化提供实验和理论依据，也对预测和诊断精神疾病以及评估脑功能有着重要意义。

Gamma rhythm(25-100Hz) in the brain is a synchronous neural activity that is commonly found in many brain regions. It is considered to be a functional feature of brain network, which is closely related to cognitive functions such as attention, memory and learning and thought to play a role in information processing, and abnormal gamma oscillation is often found to be related to mental disorders. Therefore, the understanding of the function and neural mechanism of gamma oscillation is a frontier problem in understanding the mechanism of brain information integration. However, although gamma is considered to be related to the visual cognitive function, the visual information it represents and the neural circuit mechanism are still unclear or controversial. Therefore, this paper does not focus on the cognitive function of gamma oscillation, instead, aiming at the basic neural mechanism of gamma oscillation in visual cortex, and how different neural connections of visual cortex generate and modulate gamma oscillation. The anesthetized cats were used as experimental objects, combined with electrophysiological records (in visual regions (lateral geniculate nucleus (LGN) and cortical area 18 (V1)) and mathematical modeling (large-scale neural network). The following three studies were carried out to reveal the visual response characteristics and circuit mechanism of gamma oscillation in the visual system.

Study 1 aimed at a puzzling phenomenon found by predecessors, previous studies showed that visual stimuli with high spatial frequency in the visual cortex can induce strong gamma oscillations, However, other researchers have found the opposite results in recent years. They thought that low spatial frequency visual stimulation can induce strong gamma oscillation. In order to explain this divergence, we studied how different spatial frequency information is encoded by gamma, and found a phenomenon that has never been found before: we found three different narrowband gamma oscillations in V1 for the first time, which carried different spatial frequency (SF) information, low-frequency gamma (LG, 25-40hz) tends to high spatial frequency stimulus, and medium frequency gamma (Mg, 40-65 Hz) and high frequency gamma (Hg, 65-90 Hz) tend to prefer low spatial frequency stimulus.

Study 2 further explored the neural circuit mechanisms based on the three different gamma oscillations found in Study 1. Previous studies have suggested that there are two possible origins of gamma oscillation in the visual system. One is that gamma is generated in the subcortex (including retina and lateral geniculate nucleus); The other one is that gamma is generated in the primary visual cortex. Through the layer analysis of V1, we found that the three gamma oscillations have different spatial distributions. LG is the strongest in the superficial layer, HG is the strongest in the input layer, and MG falls in between. The strongest power of HG is in the input layer suggests that it may have a subcortical origin mechanism. Then, by simultaneously recording the neural responses of the primary visual cortex (V1) and the lateral geniculate nucleus (LGN), we found that the high-frequency gamma oscillation (HG) originates from the subcortical regions, while the middle frequency (MG) and low-frequency (LG) gamma oscillation are from the cortex.

Study 3 further explored the how medium and low gamma oscillations originated in the visual cortex. There are rich neural connection types in the visual cortex, but how these neural connections affect the generation and modulation of gamma oscillation is not completely clear. In order to answer this question, we established a large-scale neural network model to simulate the primary visual cortex (V1). This mathematical model has biologically reasonable connection modes in V1: local recurrent connection (RC), feedforward connection (FF), feedback connection (FB) and long-range horizontal connection (HC). The results showed that the medium gamma is generated from the local recurrent connection, while the low gamma is generated by the long-range horizontal connection, and they are both modulated by the feedback connection.

In summary, our research on the mechanism of gamma oscillation integrated many different experimental results of predecessors (the response characteristics of spatial frequency and the origin of neural circuits of different gamma oscillations). These findings showed that gamma oscillations could reflect the neurodynamics of neural circuits, which carried different spatial frequency information in the visual system. These results not only help us deeply understand the computational principle of biological brain information coding and integration, but also have important guiding significance for understanding the neural circuit structure and cognitive function of human brain. They also provided experimental and theoretical basis for the optimization of brain like computing, and showed important significance for predicting and diagnosing mental diseases and evaluating brain function.