Neuronal calculus for the auditory pathway

TL;DR

This paper introduces a mathematical framework for modeling the early auditory pathway, emphasizing timing and connectivity, and provides analytical solutions for refractory neuron responses to periodic sounds.

Contribution

It develops a novel mathematical approach for modeling peripheral auditory neurons, including refractory and integrator neurons, with analytical formulas and numerical schemes.

Findings

Refractory neurons respond to periodic signals with asymptotically periodic output.

The model accurately reproduces basic neural activity patterns in the auditory pathway.

A numerical scheme with geometric convergence is proposed for fixed point computation.

Abstract

The first steps in the neural processing of sound are located in the auditory nerve and in the cochlear nuclei. To model the signal processing efficiently, we propose a simple mathematical tool that takes the minute timing of the system into account. In contrast to the situation in the cortex, the number of connections between neurons in auditory periphery is comparatively low. This gives way to an accurate modeling of the connectivity of the neuronal network. The timing is the all important feature in the peripheral neuronal auditory pathway. The primary auditory neurons e.g. phase lock to periodic sounds with important interactions with respect to both the refractory periods of the neurons and to the time delays caused by traveling times along the basilar membrane or through a synaptic connection. The mathematical tools provide a solid basis to build models for peripheral auditory processes. In particular, we study carefully a large class of refractory neurons, find analytical formulas for the spiking activity, and prove that refractory neurons respond to periodic signals by asymptotically periodic output. The methods rely on the theory of positive operators and give a numerical scheme for finding fixed points to an integral operator with geometric convergence rate. In addition, we consider a perfect integrator neuron, mathematically equivalent to randomized random walk, where the random walk is bounded from below, and solve the first passage time problem using continuous time Markov chain techniques. In an accompanying paper we set up the simulation framework as a counterpart to the present mathematical model. By suitably adjusting the few parameters in the model it is possible to reproduce the basic patterns of neural activity.