Functional identification of retinal ganglion cells based on neural population responses

Abstract

The issue of classification has long been a central topic in the analysis of multielectrode data, either for spike sorting or for getting insight into interactions among ensembles of neurons. Related to coding, many multivariate statistical techniques such as linear discriminant analysis (LDA) or artificial neural networks (ANN) have been used for dealing with the classification problem providing very similar performances. This is, there is no method that stands out from others and the right decision about which one to use is mainly depending on the particular cases demands. In this paper, we found groups of rabbit ganglion cells with distinguishable coding performances by means of a simple based on behaviour method. The method consisted of creating population subsets based on the autocorrelograms of the cells and grouping them according to a minimal Euclidian distance. These subpopulations shared functional properties and may be used for functional identification of the subgroups. Information theory (IT) has been used to quantify the coding capability of every subpopulation. It has been described that all cells that belonged to a certain subpopulation showed very small variances in the information they conveyed while these values were significantly different across subpopulations, suggesting that the functional separation worked around the capacity of each cell to code different stimuli. In addition, the overall informational ability of each of the generated subpopulations kept similar. This trend was present for an increasing number of classes until a critical value was reached, proposing a natural value for functional classes. © Springer-Verlag Berlin Heidelberg 2007.