Real-Time Localization of Cochlear-Implant Electrode Arrays Using Bipolar Impedance Sensing

Objective: Surgeons have no direct objective feedback on cochlear-implant electrode array (EA) positioning during insertion, yet optimal hearing outcomes are contingent on placing the EA as close as feasible to viable neural endings. This paper describes a system to non-invasively determine intracochlear positioning of an EA, without requiring any modifications to existing commercial EAs themselves. Methods: Electrical impedance has been suggested as a way to measure EA proximity to the inner wall of the cochlea that houses auditory nerve endings-the modiolus. In this paper, we extend prior work and demonstrate for the first time the relationship between bipolar access resistance and proximity of the EA to the modiolus (E-M proximity). We also evaluate two methods for producing direct, real-time estimates of E-M proximity from bipolar impedance measurements. Results: We show that bipolar access resistance is highly correlated with E-M proximity and can be approximately modeled by a power law function. This one dimensional model is shown to be capable of producing accurate real-time estimates of E-M proximity, but its simplicity also limits the potential for future improvement. To address this challenge, we propose a new prediction approach based on a recurrent neural network, which generated an overall prediction accuracy of 93.7%. Conclusion: Bipolar access resistance is highly correlated with E-M proximity, and can be used to estimate EA positioning. Significance: This work shows how impedance sensing can be used to localize an EA during insertion into the small, enclosed cochlear environment, without requiring any modifications to existing clinically used EAs.

Automated summary

This paper presents a non-invasive method to determine the positioning of a cochlear-implant electrode array (EA) in the inner ear. The study shows that bipolar access resistance is highly correlated with the proximity of the EA to the modiolus, and proposes a new prediction approach based on a recurrent neural network to improve the accuracy of estimating the EA's positioning.