Stimulus reconstruction from neural spike trains: Are conventional filters suitable for both periodic and aperiodic stimuli?

Human neurone system encodes all stimuli into series of spike trains and our brain reconstructs the stimulus back from the spikes. Main purpose of this paper is to identify the suitability of conventional filters to mimic the process of original stimulus reconstruction from neural spike trains as brain does. As human brain receives periodic and aperiodic types of signals, in this paper we have used pulse oximetry waveforms (periodic) and Gaussian signal (aperiodic) as the stimuli. For the neural spike train generation two different neurone models have been used, one model was very simple single threshold level crossing detector and the other one was an advanced simulated stochastic leaky integrate-and-fire neurone model with dynamical threshold. Level crossing detector model is used to generate spike trains from periodic signal whereas both neurone models have been used to generate spike trains for aperiodic Gaussian signal. A simple low-pass Butterworth filter and an advanced Wiener–Kolmogorov filter have been used for reconstructing the stimuli from spike trains. Comparison of the results has been done by the measure of cross correlation coefficients between the actual stimulus and the reconstructed counterpart. Comparison of the results reveal that simple level crossing detector model along with an advanced Wiener–Kolmogorov filter can achieve reconstruction of periodic signal up to 98.91% whereas combination of the advanced simulated neurone models with an advanced Wiener–Kolmogorov filter does not achieve reconstruction of more than 55% for aperiodic signal. This study proves that conventional filters are good for periodic signal reconstruction from their neural spike trains but they are not suitable for aperiodic signals.