Purchase this article with an account.
M. Bongard, J. Ammermueller, G. Ortega, P. Bonomini, E. Fernandez; Common Feature Characteristics of Ganglion Cell Populations Based on Intrinsic Deterministic Features of Spike Trains . Invest. Ophthalmol. Vis. Sci. 2003;44(13):5188.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: The encoding of single aspects of the outside world by the brain is still largely unsolved. Due to the highly parallel processing of the information transmitted from the retina to the visual centers of the brain, and the distinct large variability of single neuron responses, we investigate the encoding of visual stimuli in the responses of retinal ganglion cell populations. Methods: Responses of several retinal ganglion cell populations to visual stimuli with different physical properties were recorded using a multi-electrode array. The population activity was analyzed by looking at the correlation of synchronized appearing spikes (CSE) to reveal the deterministic behavior of the spike trains. CSE served as spike train intrinsic time reference on which all further analysis of the data in the time domain was based. The members of the different populations were classified by introducing a protocol for the identification of "temporal response classes" (TRCs). Identical TRCs of all experiments were identified according to simple statistical characteristics. The populations of different experiments were then described according to their composition of members of different TRCs. By taking this information in account we estimated the visual input from the response patterns of the retinal ganglion cell populations through artificial neuronal network analysis. Results: Estimation of the visual stimulus from the neuronal population recordings performed very well with 82%-95% correctly estimated stimuli. The exchange of members of the same TCR in the population recordings of different experiments is slightly decreasing this classification performance (70-84% correct stimulus estimation). This can be explained by the eventually lower number of spikes and therefore information in the "composed" populations. By exchanging members of different TCRs, mixing TCR data of different species, or introducing nonsense data, classification is breaking down to 8%-15% correct estimation of the stimulus used. Conclusions: The introduction of TCRs as a descriptor of the population members allows an artificial neuronal network based highly accurate estimation of the stimulus from the recorded population responses. It allows the application and successful transfer of population analysis results from one experiment to another, and therefore provides a way to extract common features in the responses of different retinal ganglion cell populations.
This PDF is available to Subscribers Only