Signal Processing For Multi-Electrode Recordings From Retinal Ganglion Cells

The first stage of visual information processing occurs in the retina. Visual stimuli excite the photoreceptors, which convert the visual stimuli into electrical signals. These signals are then propagated through the axons to the output neurons - the ganglion cells and are relayed to the brain in the form of the action potentials. A primary goal of retinal research is to determine how visual information is encoded and transmitted in the retinal circuitry. This is not only to understand the biology system, but also to provide the biology bases for the development and application of the artificial vision. In order to understand the information processing in retinal ganglion cells, the recently developed multi-electrode arrays have been employed. However, this technique challenges the biology signal processing. One the one hand, the signal-to-noise ratio of multi-unit extracellular recordings is low, and the propertiesb of frequency spectrum between the signal and the noise are similar. At the same time, the multi-unit extracellular recordings contain overlapped spikes produced by a number of neurons adjacent to the electrode and spikes from one neuron may be recorded by several electrodes. On the other hand, it becomes clear that more often than not, neural information is processed based on cooperation and integration of relevant neuron populations. However, the ability of understanding relevant coding function is significantly hampered because current methods fall short of the requirements for relevant multi-dimensional data analysis.The innovations in the paper as follows:Firstly, the methods for spike detection and spike sorting were discussed and researched based on the spike characters of freshly hatched chicken's ganglion cells. A robust method for spike sorting with automatic overlap decomposition was proposed. The higher effectiveness of the method is validated by synthetic data and the new method is robust to background noise in real application for no assumptions about noise properties are requested.Secondly, nonlinear methods were used to analyze the spike trains. In the research, nonlinear predictability was expected in the process of spontaneous firing of ganglion cells by employing the nonlinear forecasting method with surrogate data analysis, which implied that the process was chaos one. Further, the difference in temporal structures of retinal ganglion cell spike trains between spontaneous activity and firing activity after contrast adaptation was investigated. The result implied that the behavior of the neuron became ordered, which might carry relevant information about the external stimulus and external information could be encoded in forms of some certain patterns in the temporal structure of spike train.Finally, the application of information theory in multi-sequence was discussed and a new method of multiple dimension data processing based on a measurement of information discrepancy was researched. Moreover, the new method was employed to deal with the multiple spike trains simultaneously recorded from a group of neurons and analyze the space-time pattern among these neurons, and the efficiency and veracity of the method were validated in real application.