Use of computational modeling of auditory nerve activation patterns for the optimization of cochlear implant electrical stimulation patterns

Evaluation of user-independent efficacy of novel cochlear implant (CI) signal processing strategies is difficult. Current practice relies on customization based on user performance in the clinical setting. As a result, it is impractical to achieve theoretical optimality, and improvements to stimulation strategies are frequently achieved using trial-and-error. A computational framework that approximates the upper bound on CI performance is desirable to reduce the number of novel strategies that require evaluation in patients. Therefore, the goal is a physiologically-based framework, using accepted models of the acoustically- and electrically-induced nerve activation patterns (NAPs), to predict discrimination and forced-choice identification among acoustic stimuli. An initial framework has been developed to predict behavioral performance of normal-hearing (NH) listeners, using NAPs generated from the Zilany-Bruce auditory-periphery model. The correlation-based metric predicts rank orders of errors in confusion matrices exhibited by NH listeners during psychophysical experiments. This method of predicting behavioral differences between stimuli has been extended to a framework that generates the optimal electrode stimulation sequence on an individual subject and stimulus basis by using the Bruce neural model for electrical stimulation. While generating these optimal electrical firing patterns is not feasible for use in real-time CI signal processing, it provides an estimate of the upper bound on CI user performance and allows for future work in designing real-time signal processing algorithms