An information theoretic approach for optimizing the analysis of multi-electrode multi-unit neuronal recordings

This paper introduces an information theoretic approach for reducing the dimension of the observation space of multiunit neuronal responses recorded by high-density microelectrode arrays in the nervous system. The technique relies on selectively processing electrode channels that record neuronal responses to common information content without actually sorting the spikes. It exploits the statistical dependency of the data by maximizing the likelihood of observing covariance structures given a set of distance measures of the mutual information between the selected data channels. The strength of the proposed technique relies on the ability to measure mutual information of mixtures of neuronal responses to allow the selectivity to take place. Our preliminary results demonstrate that appropriate selection of data channels sharing common information prior to sorting the individual neuronal spike trains is feasible. Accordingly, substantial savings in computational costs and minimized user intervention are obtained compared to raw data transmission for off-line analysis.