An empirically based model of the electrically evoked compound action potential

The relationship between electrically evoked single-fiber action potentials and the electrically evoked compound action potential of the auditory nerve is of interest to those attempting to model such responses with computational techniques. It also relates to efforts to exploit the gross potentials that can now be recorded by some implantable cochlear prostheses. In this paper, we develop a computational model of the auditory nerve response to single, pulsatile, electrical stimuli based upon the response characteristics obtained from 230 single fibers of 13 cats. These fibers were stimulated by brief (39s) monophasic cathodic stimuli delivered by a monopolar intracochlear electrode. The data were pooled to obtain an estimate of the distribution of fiber thresholds. Post-stimulus time histograms were modeled using Poisson functions and adjusted to account for empirically determined latency and jitter characteristics. The probabilistic nature of single-fiber input-output functions (i.e. Verveen's (1961) 'relative spread') was also modelled. PST histograms from 5000 modelled fibers were then summed and convolved with an estimated 'unit potential' following the method of Goldstein and Kiang (1958). This convolution produced modelled compound action potentials, which were then compared with experimentally obtained data. Manipulations of model parameters affecting threshold, jitter, and relative spread suggest that the most important determinant of the shape of the EAP amplitude-level function is the threshold distribution. A model based solely on threshold distribution produces an EAP input-output function similar to one that accounts for probabilistic single-fiber input-output functions. Discrepancies between these two models do occur if the threshold distribution function is compressed significantly, as might be the case in pathological cochleae with altered distributions or numbers of nerve fibers.