Search Papers | Poster Sessions | All Posters
Poster B108 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Neural Network Models of Hearing Clarify Factors Limiting Cochlear Implant Outcomes
Annesya Banerjee1,2 (), Mark R. Saddler2, Josh H. McDermott2; 1Harvard University, 2MIT
Cochlear implants (CIs) allow deaf individuals to hear by electrically stimulating the auditory nerve, bypassing the ear. CIs are one of the great successes of biomedical engineering, but nonetheless fail to restore normal auditory perception. Models that can predict behavioral outcomes given CI input could help diagnose the factors limiting perception and thus guide device improvements. We first built a model of normal hearing by optimizing a deep neural network to perform real-world auditory tasks using simulated auditory nerve input from an intact cochlea. We then modeled CI hearing by testing this model on simulated auditory nerve responses to CI stimulation. To simulate possible consequences of learning to hear through a CI, we re-optimized the network on CI input. When the entire network was reoptimized, the model exhibited speech intelligibility scores significantly better than typical CI users. Speech recognition on par with typical CI users was achieved only when just the late stages of the model were reoptimized. However, sound localization performance remained abnormal relative to normal hearing even when the entire network was reoptimized for CI input. The results suggest that some limitations of CIs reflect impoverished peripheral information from potentially suboptimal stimulation strategies, but that other limitations may reflect incomplete central plasticity.
Keywords: Brain plasticity Cochlear implants Neural networks Auditory perception