【作者】 张莹莹；

【导师】 梁培基；

【作者基本信息】 上海交通大学 ， 生物医学工程， 2011， 博士

【摘要】 视网膜是视觉信息处理的第一站,视网膜光感受器细胞接受视觉信号输入并将其以膜电位改变的形式经视网膜神经元回路传递至输出神经元神经节细胞,形成动作电位,然后通过视神经进一步向视觉中枢系统传递。视觉经验通常都发生在自然环境中,因此直接使用自然图像刺激来探索视觉系统中某些未知的特性越来越受到视觉系统研究者的重视。视网膜的信息处理过程受到视神经在视觉通路解剖结构中的瓶颈地位,以及神经系统代谢能量消耗等诸多因素的制约。本研究的目的就是通过分析群体视网膜神经节细胞在连续自然影像(电影)刺激下反应活动的时空特性来了解视网膜是如何在这些限制下有效地对视觉信息进行处理和编码的。本论文运用多电极记录技术结合数据分析,以伪随机棋盘格(pseudorandom checker-board)刺激为对照,对小鸡视网膜神经节细胞在自然刺激下的时空编码模式进行研究,包括考察神经节细胞群体在自然影像刺激下放电活动在时间上和空间上的统计特性,以及这种统计特性下群体神经节细胞的放电模式和协同作用等。主要结果包括:1)在自然影像的刺激下,单个神经元在刺激延续过程中的放电序列以及群体神经元在特定时刻的反应活动呈现出比在伪随机棋盘格刺激下更明显的超高斯分布,提示群体神经节细胞可能以一种稀疏的反应模式编码自然视觉刺激信息。2)进一步分析研究发现,这种稀疏编码模式的形成更多地依赖单个神经节细胞在刺激延续过程中的少数高频放电和多个神经元在某一时刻的协同放电,而不仅仅是在刺激过程中大多数神经元在大多数时间里不放电。这些结果提示视网膜神经节细胞在自然视觉信息处理过程中可能通过单个神经节细胞的少数高频放电和群体神经元中邻近神经元的协同放电实现视觉信息稳定而高效地向视觉中枢传递。3)应用基于子序列分布差异度量(measurement of sub-sequence distribution discrepancy, MSDD)的多维数据处理方法分析群体神经节细胞在不同刺激时段放电活动的协同性,发现在自然影像刺激下群体神经节细胞在刺激过程中的协同放电随时间存在明显的动态特性,进一步采用基于熵分析的信息论算法对群体神经节细胞在自然刺激下的完整放电序列进行分析以考察群体神经节细胞中不同成员在整个刺激过程中动态成组协同放电的情况,发现在自然刺激下,群体神经节细胞的协同放电模式比在伪随机棋盘格刺激下更加广泛和多样化,提示神经节细胞在自然环境中以一种更加高效的方式进行信息传递。从这些结果可以看出,视网膜可以通过群体神经节细胞的时空发放模式来对信息进行编码,这种模式不仅节省了储存大量信息所必需的神经元数目,节省了代谢能量消耗,有效地实现了信息之间的快速传递,同时也加大了对外界信号的识别和处理能力。而视网膜群体神经节细胞的这种以单个神经元在少数时刻的高频放电及相邻神经元动态成组协同放电为特征的高效反应模式(即稀疏编码)在对自然影像刺激的编码中体现得尤为突出,更有力地说明了视网膜神经节细胞在自然环境下具有更优化的时空编码特性。更多还原

【Abstract】 The first stage of the visual information processing occurs in the retina. The photoreceptors absorb photons from the visual field, and signal the light information through changes in their membrane potentials. These signals are further transmitted along the retinal circuit to the output neurons, retinal ganglion cells and make their way to the central visual system in the form of action potentials (spikes). Nearly all the visual experiences happen in natural environment, thus it has been paid more and more attention to investigate the visual information processing using natural stimuli. The information processing in the retina is subject to several restrictions such as the anatomic bottleneck structure of the retina in the visual pathway and the limited metabolic energy. The aim of this investigation is getting to know how the retinal ganglion cells process and encode the natural visual stimuli effectively under such constraints. In the present work, the multi-neuronal activities of the chick retinal ganglion cells in response to time-varying natural images (movies) as well as pseudorandom white-noise checker-board flickering sequence (control) were recorded using multi-electrode recording technique. The statistical distributions of the population retinal ganglion cells’spikes in temporal (over the duration of stimulation) and spatial (across the population neurons), the underlying firing patterns and dynamic grouping of activated neurons in synchronization, were analyzed and investigated.The main findings include three parts: 1) In response to the natural stimuli, the individual single retinal ganglion cell’s firing activities over the duration of the stimulation and the multi-neuronal instantaneous activities at the same moment both show more obvious super-Gaussian distribution, which is related to sparse coding of the neurons, than those during stimulation of pseudorandom white-noise checker-board flickering sequence. 2) Further analysis of the experimental data shows that it is the action potentials fired in“burst”form of individual single neurons and synchronized activities of neurons in group that contribute mainly to the sparse representation of the population retinal ganglion cells. These results suggest that in response to natural stimuli, individual neuron fires at a low rate to save metabolic energy, while the dynamically grouped small subsets of neurons are activated with adjacent neurons firing concertedly and individual single neurons fire in“burst”form occasionally to transmit information to the postsynaptic neurons efficiently. 3) Multi-dimensional data analysis based on MSDD (a measurement based on sub-sequence distribution discrepancy) was applied to the spike trains of population retinal ganglion cell during different time periods of the stimulation and the results show that the neurons fried synchronously in dynamic groups over the duration of natural stimulation. Furthermore, an information-theoretic algorithm based on entropy analysis was applied to identify the synchronized neuronal groups of the population RGCs based on the complete spike trains over stimulation. The results show that the synchronized firings in RGCs are more extensive and diverse and may account for more effective information processing in representing the natural visual environment.These results indicate that the retina processes the natural visual information in a highly dynamic and efficient manner through the spatiotemporal firing patterns of the population ganglion cells, which can reduce the number of neurons involved, save the metabolic energy consuming, improve the efficacy of the information transmission and enhance capabilities of the ganglion cells’information reorganization and processing.更多还原