Optimization of Cochlear Implant Electrode Selection Strategy Enhances Vowel Identification in Noise

Introduction: While adults with post-lingual hearing loss generally achieve excellent results with cochlear implantation, there is large variability in both audiologic performance and patient-reported quality of life (Fabie et al., 2023). Some patients are unable to attain desired listening outcomes despite aural rehabilitation and standard clinical adjustments to programming, and both biological and device-related factors have been associated with unsatisfactory performance (Moberly et al., 2016). We seek to understand whether a unique, auditory nerve-derived electrode selection strategy may improve performance for cochlear implant (CI) patients exhibiting suboptimal performance. This effort builds on two previous pilots in which a Nucleus-24 implant was directly driven using our novel electrode selection strategy (Aguiar et al., ARO 2013; Llico et al., BMES 2017). In both cases, patient performance in a closed-set vowel identification (/hVd/) task was improved, in an absolute sense, by at least 10% when stimuli using the novel electrode selection strategy were presented in noise with a 10 dB SNR. The goal of this study is to systematically assess this novel programming strategy in a larger population to confirm the performance boost and to enable continued translational development of this electrode selection strategy.

Materials and

Methods: Electrode Selection Framework: Optimized electrode selection sequences are generated as follows. First, computational models of the auditory peripheral system (Aguiar, 2012; Llico, 2021) are used to approximate the neural activation pattern (NAP) elicited in the auditory nerve in response to an acoustic stimulation under normal hearing (NH) conditions. This NH-NAP is then used as the target of an optimization process (Lee and Allebach, 2002) to which the CI-NAP induced from electric stimulation by the CI electrodes (Bruce et al., 1999) is to be matched. Note that this current optimization strategy is slow, leading to a six hour computation time to optimize the stimulation for a half-second utterance.
Subjects: Three post-lingually deafened CI users have participated in this study to-date, with each undergoing at least two one-hour sessions in which they complete 100 trials of a closed-set task to identify vowels (/hVd/). Within each session, the task was performed twice, once using source stimuli without noise (infinite SNR), and once using source stimuli in speech-shaped noise at 10 dB SNR. Full participation will involve four sessions conducted in an A-B-A-B paradigm, varying the electrode selection strategy by session, either using the normal manufacturer settings ("baseline") or the optimized strategy.

Results, Conclusions, and Discussions: Preliminary results indicate that there is an increased performance in vowel identification rates when using the optimized selection strategy, relative to the patient's normal CI settings. Subject CI000 completed all four sessions. They achieved an identification rate of 80% for both strategies at infinite SNR, but at 10 dB SNR their baseline performance was 68% while the optimized strategy again yielded 80% performance. Preliminary analysis of the two sessions of data collected for Subject CI001 yields identification rates of 36% at infinite SNR and 20% at 10 dB SNR for their baseline, and 56% and 39%, respectively, for the optimized strategy. Similarly, subject CI002 achieved identification rates of 24% (infinite SNR) and 28% (10 dB SNR) for baseline, and 39% and 33%, respectively, for the optimized strategy. Across the three participants, vowel identification using the default manufacturer electrode selection strategy is 46% at infinite SNR and 39% at 10 dB SNR (44% overall), whereas the optimized electrode selection strategy produced an identification rates of 58% at infinite SNR and 51% at 10 dB SNR (55% overall).
It is hypothesized that the regular stimulation patterns associated with normal commercial CI electrode stimulation strategies (e.g., repeated sequential stimulation of all or a subset of electrodes) can produce the beating of multiple periodic structures, a behavior known from digital imaging to produce Moire patterns (Baqai et al., 2005). If such Moire patterns are present in the NAP induced by the CI, users may find it more difficult to interpret incoming signals as such patterns could be interpreted by the central auditory pathway as (spurious) pitch contours. The optimized electrode selection strategy implemented in this framework will produce a more stochastic electrode stimulation, with such a pattern known to reduce the perception of Moire patterns (Baqai et al., 2005).
Should additional data collection continue to evidence a robust gain from use of the optimized electrode selection strategy, future effort will be directed at development of rapid approximations of the current approach, targeting achievement of a near-optimal electrode selection within 25 ms.

Acknowledgements (Optional):