【作者】 蔡超峰；

【导师】 梁培基； 张溥明；

【作者基本信息】 上海交通大学 ， 生物医学工程， 2009， 博士

【摘要】 视觉信号的初级处理发生在视网膜。视觉刺激被光感受器细胞转换成电信号,经由视网膜神经回路传递至视网膜神经节细胞,然后通过视神经进一步向视觉中枢传递。作为视网膜的输出神经元,神经节细胞发放的动作电位序列编码了视觉刺激所携带的信息。因此,对视网膜神经节细胞放电活动的研究有助于了解视网膜的信息处理和编码机制。此类研究的意义不仅在于对生物系统的了解,同时也为人工视觉的开发和利用提供生物学上的依据。使用自然刺激对探索视网膜如何对自然环境下的视觉信息进行处理和编码极为重要。然而自然刺激的统计特性非常复杂,神经系统本身还是一个复杂的非线性系统,这就使得视网膜神经节细胞在自然刺激下的放电活动往往表现出非周期的、不规则的形式。在分析视网膜神经节细胞这种非周期的、不规则的放电活动时,首先应该判断其中是否包含确定性,从而为下一步的数据分析工作奠定基础。论文第一部分在单细胞层面上采用关联维数法和非线性预测法对视网膜神经节细胞在自然电影刺激和棋盘格刺激下的放电活动序列进行了分析。分析结果包含三部分的内容。第一,视网膜神经节细胞不规则的、非周期的放电活动包含着内在的确定性,因此我们可以采用确定性模型对神经节细胞的放电活动进行研究。第二,对视网膜神经节细胞放电活动时间间隔序列的差分有利于确定性的检测。第三,在检测时间序列是否包含确定性时,非线性预测法比关联维数法有效。实验发现,一些小鸡视网膜神经节细胞在重复性全域闪光刺激下的放电活动呈现出一种独特的时间结构特征—双峰响应。目前,这种双峰响应的产生机制还不清楚。论文第二部分在视网膜层面上通过建立视网膜模型的方法对双峰响应的产生机制问题进行了研究。模型分析结果表明,受到慢瞬时给光/撤光型无长突细胞抑制的快瞬时给光/撤光型无长突细胞对持续型神经节细胞产生抑制作用,使得神经节细胞持续型响应中特定时间段内的放电活动受到抑制,从而形成双峰响应。分析结果提示,无长突细胞对神经节细胞放电活动的调制作用使得神经节细胞的放电活动表现出了丰富的时间结构特征。高斯差模型描述了视网膜神经节细胞经典的同心圆式中心—外周拮抗的感受野结构特征。后来的研究发现,视网膜神经节细胞经典感受野之外还存在着一个范围很大的去抑制区,这个区域能够削弱经典感受野内外周区对中心区的抑制性作用。去抑制模型解释了这种去抑制作用内在的产生机制。论文第三部分通过建立视觉系统模型的方法对这两种感受野模型的编码质量问题进行了研究。视觉系统模型包含一个视网膜模块和一个中枢视觉系统模块,视网膜模块由一组空时滤波器组成,每个空时滤波器模拟一个视网膜神经节细胞。计算结果表明,当视网膜模块中的空时滤波器采用去抑制模型实现时,中枢视觉系统模块能够从视网膜模块所产生的动作电位序列中提取出更多的信息,亦即去抑制模型的编码质量较高。由此可见,经典感受野之外的去抑制区对于视网膜神经节细胞的信息编码具有重要意义。这为图像处理技术的发展以及人工视觉的开发和利用提供了参考。更多还原

【Abstract】 The initial visual information processing occurs in the retina. Photoreceptors convert the visual stimuli into electrical signals. These signals are propagated through the retinal circuit to the ganglion cells and then to the central visual system. As the output neurons of the retina, ganglion cells fire action potential trains which encode the visual information carried by the visual stimuli. Therefore, study on the firing activities of retinal ganglion cells helps to understand the information processing and encoding mechanism of the retina. This not only contributes to the understanding of the biology system, but also provides biology bases for the development and application of the artificial vision.Natural stimuli are important to explore the information processing and encoding mechanism of the retina under natural scenes. However, due to the complicated statistics of natural stimuli and the nonlinearity of neural system, the firing activities of retinal ganglion cells during response to natural stimuli are always aperiodic and irregular. Determinism detection in the aperiodic, irregular firing activities of retinal ganglion cells provides directions for future work. In the first part of this dissertation, the correlation dimension method and the nonlinear forecasting method were applied to analyze the firing activities of retinal ganglion cells during response to natural movie and checker-board flickering. Mainly three conclusions were drawn. Firstly, there exists determinism in the aperiodic, irregular firing activities of retinal ganglion cells. As a result, deterministic models should be employed to study the firing activities of retinal ganglion cells. Secondly, the difference operation of the inter-spike interval series makes it easier for determinism detection in the firing activities of retinal ganglion cells. Thirdly, the nonlinear forecasting method is more efficient than the correlation dimension method for determinism detection.In the experiment with repetitive light flashes, it was found that the firing activities of a subpopulation of chicken retinal ganglion cells show a particular temporal structure– dual-peak response. The generation mechanism of the dual-peak response remains unclear. In the second part of this dissertation, a retina model was developed to solve this question. Model results indicate that the dual-peak response results from an inhibitory input to a sustained ganglion cell from a fast transient ON/OFF amacrine cell inhibited by a slow transient ON/OFF amacrine cell, which inhibit the ganglion cell’s sustained response at a certain time window. These results imply that the firing activities of ganglion cells are regulated by the amacrine cells and thus demonstrate rich temporal structure.The difference of Gaussian model describes the classical antagonistic center-surround receptive field structure of retinal ganglion cells. However, it has been reported that there exists an extensive disinhibitory region beyond the classical receptive field and this disinhibitory region attenuates the surround inhibition in the classical receptive field. The disinhibition model explains the intrinsic mechanism of the disinhibition. In the last part of this dissertation, a visual system model was developed to study the encoding quality of the two receptive field models. The visual system model includes a retinal module and a central visual system module. The retina module consists of an ensemble of spatial-temporal filters and each filter simulates an individual ganglion cell. Model results indicate that the central visual system module decodes more information from the action potential trains generated by the retina module when the spatial-temporal filters are implemented by the disinhibition model. Therefore, a conclusion was drawn that the encoding quality of the disinhibition model excels the difference of Gaussian model. These results imply that the disinhibitory region beyond the classical receptive field is of great significance to the information encoding of retinal ganglion cells. This provides reference for the development of image processing and application of artificial vision.更多还原