The Neural Coding of Retinathe Computational Models of Two Types W Ganglion Cells in the Cat
|School||Shanghai Jiaotong University|
|Course||Control Theory and Control Engineering|
|Keywords||Cat retina Ganglion cells Bipolar cells Amacrine cells X cells Y cells W cells Local edge detectors Tonic suppressed-by-contrast cells Uniformity detectors Edge inhibitory off-center cells G17 cells Classical receptive field Non-classical receptive field Silent inhibitory surround Spatiotemporal model Recurrent network|
From the perspective of a cybernetical theorist, the neural coding of a retinal ganglion cell is that the light stimulus is processed by the transfer function of the cell, and then is relayed to the brain. Therefore, finding the neural coding of a cell is equal to finding the transfer function of the cell. From a more general view, the computational model of a cell is just its transfer function.W cells play an essential role in the visual system of the cat. W cells project to a variety of cortical targets such as the C-laminae of the lateral geniculate nucleus, the medial interlaminar nucleus, the superior colliculus, the pulvinar, and the nuclei of the accessory optic system. These nuclei further relay the signals to cortical areas 17 (the primary visual cortex), 18, and 19. Moreover, these three cortical areas are presumably involved in eye movement, pupillary control, and motion perception. Therefore, a computational model of W cells will be beneficial in providing a better understanding of the functions of W cells, the retina, and the visual system.Although W cells play a critical role in the visual system, in the literature, little attention has been given to explain how these cells work. In this study, the computational models are proposed to describe two typical types of cat W cells, a local edge detector, and a suppressed-by-contrast cell.Based on the electrical coupling between retinal cells, the conception of‘feedback’is introduced, and two recurrent network models are used to describe the behaviors of the impressed-by-contrast cell and the suppressed-by-contrast cell in the cat retina. An additive recurrent network is used to characterize the impressed-by-contrast cell, and the other additive recurrent network with saturation rectification is used to characterize the suppressed-by-contrast cell. Through simulations, it is found that recurrent networks are able to describe qualitatively the responses of these two types of W cells to drifting sinusoidal gratings.The recurrent network model is partly supported by the anatomical structure of the retina, while the traditional feedforward model is more closely related to the structure of the retina. Therefore, feedforward models are used to fully characterize these two types of W cells.A feedforward spatiotemporal model is proposed for local edge detector cells (impressed-by contrast cells) in the cat retina. The model is able to describe these cells’ responses to drifting sinusoidal gratings, and these responses are consistent with the physiological observations. The model is also able to predict their responses to alternating sinusoidal gratings, flashing spots and annuli, and these predictions are qualitatively close to the experimental findings. The anatomical implement of the model is discussed in detail, and it is found that each component of the model may be fulfilled by the specific retinal cell. Therefore, the model is physiologically plausible. In conclusions, the model qualitatively summarizes these cells’ behaviors, and may be useful in explaining local edge detector cells in other vertebrates.The physiological data on tonic suppressed-by-contrast cells is more infrequently reported than local edge detector cells, and thus a mathematical model of tonic suppressed-by-contrast cells is more valuable than a model of local edge detector cells.A spatiotemporal model composed of two pathways, an edge pathway and a linear pathway, is introduced to describe the behaviors of tonic suppressed-by-contrast cells in the cat retina. The edge pathway maps the cells’edge inhibitory behaviors, and the linear pathway maps their off-center concentric behaviors. The model qualitatively summarizes the cells’responses to sinusoidal gratings and flashing spots. The organizations of the model map well the anatomical retinal circuitry of the cat, and the model may be useful in explaining the responses of the suppressed-by-contrast cells of other species.The non-classical receptive field of a ganglion cell also plays a fundamental role in the visual system of the cat. It may help the retina transmit information about the area brightness and gray scale, and may also help a ganglion cell to tune dynamically the parameters of its classical receptive field. Its orientation selection may permit the retina to detect the sophisticated natural image, and it may also contribute a lot to the second-order process of a cortical cell. Therefore, a computational model of the non-classical receptive field of a ganglion cell will help us get a better understanding of the functions of a ganglion cell.A theoretical model of cat X cells is proposed, and it integrates the X cell’s classical center surround receptive field mechanism with its non-classical receptive field mechanism. The model is able to simulate the spatial and temporal frequency response curves of the classical receptive field mechanism. Moreover, it can qualitatively reproduce the spatial and temporal frequency response curve of the non-classical receptive field mechanism. Finally, the model can describe the interaction between the classical receptive field mechanism and the non-classical receptive field mechanism. It is suggested that the non-classical receptive field mechanism of retinal ganglion cells might be shaped from the feedforward summation of bipolar cells and amacrine cells. The model in this study is more physiologically plausible, compared with the assumption that the non-classical receptive field mechanism of retinal ganglion cells may be formed from the mutual inhibition among amacrine cells.