Towards model-based characterization of individual electrically stimulated nerve fibers

Neuroprosthetics can partially restore the impaired system’s functionality. To improve such prosthetics, a quantitative understanding of the electrical stimulation of nerve fibers is required. This knowledge can best be represented by computational models of the process. Currently, most models of electrically stimulated nerve fibers are based on many different datasets, which mainly consist of the average analysis values of recordings of many nerve fibers. While this is a valid approach for understanding the fundamental neurophysiological response properties, both the combination of many different datasets and the average analysis can confound details in the response of the nerve fiber. To improve computational models of electrically stimulated nerve fibers further, we propose an optimization procedure that can fit the parameters of a neuron model to the response of a single nerve fiber to pulse-train stimulation. We show that in this way, the model can reproduce a wide variety of fiber responses of electrically stimulated auditory nerve fibers of guinea pigs in a remarkably detailed way on a scale of less than 1 ms. This not only illustrates the ability of the model, but also the quality and consistency of the data. We analyze and discuss the certainty and generalizability of the parameter sets thus exposed. The model parameters found by the optimization procedure can then form the basis for a detailed fiber-by-fiber analysis, which we illustrate by a correlation analysis of the predicted response properties (e.g., spike latency and refractory period) in the fiber response.Author summary: Neuroprosthetics can partially restore the function of an impaired neural system. Examples of such prosthetics are cochlear implants, which allow deaf people to hear, or deep-brain stimulators, which can reduce the tremor in Parkinson’s disease. While the mere existence of such prosthetics is already impressive, there is still room for improvement. For example, cochlear implant users have problems understanding speech in background noise, and deep-brain stimulators can have serious side effects, such as speech difficulties or limited fine motor control. To improve such implants, a detailed understanding of the underlying processes in the electrically stimulated of nerve fibers is required. A valuable representation of our understanding of the process is a computational model of electrically stimulated nerve fibers. Here, we optimized the parameters of one model, such that the behavior of 118 individual nerve fibers was represented, resulting in a single parameter set per fiber. In this way, a single model can reproduce a wide variety of response patterns, which allows for a detailed analysis of the individual fiber based on these parameters.