Decoding neural spike trains: calculating the probability that a spike train and an external signal are related.

Abstract

Experimental and clinical applications of extracellular recordings of spiking cell activity frequently are used to relate the activity of a cell to externally measurable signals such as surface potentials, sensory stimuli, or movement measurements. When the external signal is time-varying, correlation methods have traditionally been used to quantify the degree of relation with the neural firing. However, in some circumstances correlation methods can give misleading results. A new algorithm is described that estimates the extent to which a spike train is related to a continuous time-varying signal. The technique calculates the probability of generating a spike train with Poisson statistics if the time-varying signal determines the Poisson rate. This is accomplished by successive division of the signal and the spike train into halves and recursive calculation of the probability of each half-signal. The performance of the new algorithm is compared with the performance of correlation methods on simulated data.