ICF - Restoring the neural code through optimal hearing aids: Toward validated prototypes
Lead Research Organisation:
UNIVERSITY COLLEGE LONDON
Abstract
Hearing loss is one of the most widespread and disabling conditions in the world, affecting approximately 500 million people. Given the enormous individual, societal, and economic burden of hearing loss, there is urgent need for effective treatments. For the most common forms of hearing loss, there is little chance of an effective biotherapeutic in the near future; the only hope rests with hearing aids. Unfortunately, current devices perform poorly in typical social settings with loud sounds and background noise and, thus, there remains a huge unmet clinical need.
It is no surprise that hearing loss cannot be corrected by the simple signal processing in current hearing aids. Hearing loss is a complex problem that causes dramatic distortions in the information that the ear sends to the brain and alters the way in which the brain processes the information it receives. To improve performance, the next generation of hearing aids must incorporate more powerful signal processing that compensates for these changes. Designing this new signal processing is much easier said than done but, fortunately, recent advances in neuroscience and machine learning have allowed us to finally make progress.
We have taken a new data-driven approach to hearing aid design. Using recordings of brain activity with and without hearing loss, we train deep neural networks to perform the signal processing that provides the best possible compensation for the distortions caused by hearing loss. We have spent the past several years developing unique experimental methodologies for large-scale, high-resolution neural recordings, collecting the required datasets, and training neural networks (which we call AidNets) to act as optimal hearing aids. Our initial results, which demonstrate that AidNets can be far superior to the state-of-the-art, provide strong proof of concept and we have recently submitted a patent application as an initial step toward commercialization.
In this proposal, we detail our plans to develop validated prototypes. Our key aims include (1) developing a framework for personalizing AidNets for individual listeners; (2) optimizing AidNet-based software prototypes for speech-in-noise and music; and (3) benchmarking the prototypes against state-of-the-art hearing aids in human listening tests. The prototypes will be targeted toward those who are most likely to benefit from improved hearing aids – people with mild-to-moderate sensorineural hearing loss – but can be adapted to other populations in the future.
Between our core team and our collaborators, we have all of the expertise required to achieve our aims. We also have the additional support that will be required to ultimately bring our technology to market and realize patient benefit, with partners including the UCLH Royal National ENT Hospital and its associated Biomedical Research Centre; UCL Business; the UCL Translational Research Office; and our startup company, Perceptual Technologies Ltd. These partners will help us to engage stakeholders (patients, clinicians, investors, and industry) ensuring that we make the most of our opportunity to develop life-changing technologies for millions of people.
This application relates to the AI, engineering bio & quantum tech highlight notice.
It is no surprise that hearing loss cannot be corrected by the simple signal processing in current hearing aids. Hearing loss is a complex problem that causes dramatic distortions in the information that the ear sends to the brain and alters the way in which the brain processes the information it receives. To improve performance, the next generation of hearing aids must incorporate more powerful signal processing that compensates for these changes. Designing this new signal processing is much easier said than done but, fortunately, recent advances in neuroscience and machine learning have allowed us to finally make progress.
We have taken a new data-driven approach to hearing aid design. Using recordings of brain activity with and without hearing loss, we train deep neural networks to perform the signal processing that provides the best possible compensation for the distortions caused by hearing loss. We have spent the past several years developing unique experimental methodologies for large-scale, high-resolution neural recordings, collecting the required datasets, and training neural networks (which we call AidNets) to act as optimal hearing aids. Our initial results, which demonstrate that AidNets can be far superior to the state-of-the-art, provide strong proof of concept and we have recently submitted a patent application as an initial step toward commercialization.
In this proposal, we detail our plans to develop validated prototypes. Our key aims include (1) developing a framework for personalizing AidNets for individual listeners; (2) optimizing AidNet-based software prototypes for speech-in-noise and music; and (3) benchmarking the prototypes against state-of-the-art hearing aids in human listening tests. The prototypes will be targeted toward those who are most likely to benefit from improved hearing aids – people with mild-to-moderate sensorineural hearing loss – but can be adapted to other populations in the future.
Between our core team and our collaborators, we have all of the expertise required to achieve our aims. We also have the additional support that will be required to ultimately bring our technology to market and realize patient benefit, with partners including the UCLH Royal National ENT Hospital and its associated Biomedical Research Centre; UCL Business; the UCL Translational Research Office; and our startup company, Perceptual Technologies Ltd. These partners will help us to engage stakeholders (patients, clinicians, investors, and industry) ensuring that we make the most of our opportunity to develop life-changing technologies for millions of people.
This application relates to the AI, engineering bio & quantum tech highlight notice.