Introducing Data Science: Hearing aids on the brink of a paradigm shift

In the near future, hearing aid users will get the opportunity to participate in hearing aid development. They will be facilitated to fine-tune some settings by themselves. Results by hundreds of thousands of users will be submitted to big data analysis engines, aimed at algorithm optimisation. A new and exciting chapter in the history of hearing aid development is about to start. Professor Bert de Vries compares this shift to the transition from analog to digital devices that set in nearly two decades ago.

The long term trend in healthcare is a bigger role for the patient, who carries out and decides more by himself. There are several main drivers for this shift towards do-it-yourself healthcare. An important aspect is the political pressure to counter the rapidly rising healthcare costs. The second main aspect is the availability of miniaturised and wireless communication technology, the internet, wearable electronics and labs-on-a-chip, enabling the monitoring of body and mind functions and linking the results to remote analysis. A third main driver involves growing insight in individual differences. Modern biological drugs will not work for general populations, but do work for certain categories in them, depending on their specific genes or immune system. Nowadays healthcare demands a personalised approach or personalised medicine. “As a result, a 'patient-centric medical care system' is coming into existence”, concludes Bert de Vries, Ph.D. He is a part-time full professor in Personalization of Medical Signal Processing Systems at Eindhoven University of Technology (TU/e) and researcher at GN Resound in Eindhoven. “Personalisation is also the way to go forward for the hearing aid industry. It will help lower the number of return visits to the clinic or shop. More importantly, it leads the way towards higher levels of user satisfaction for hearing aid wearers.”

What is a hearing aid?

“Today’s hearing aids are technically advanced devices, which combine the paradoxical demands of complex audio processing with very low energy consumption”, De Vries states. “The audio processing is complex, because next to compensating for hearing loss, the hearing device has to deal with challenging acoustical problems in and around the ear that were introduced by the device itself, such as feedback, occlusion, loss of directional hearing and so on.” All this has to be taken care of before the restoration of hearing loss can even begin. De Vries: “To compensate the hearing loss is also challenging, because we don't know exactly how the processing in the brain works. Neuropathological knowledge has improved significantly lately, but this refers more to qualitative insights than the quantitative models that we signal processing researchers need to support our mathematical approach to algorithm design.” It is known that the brain is highly selective and throws away lots of redundant and useless information early in the auditory pathway. It selects on the basis of expectation: hearing is a fusion of predictions and actual sensory signals. The complexity of a hearing aid lies in that it has to restore hearing loss in support of this fusion process. De Vries: “A hearing aid is basically an algorithm that alters the sound to which the patient listens, with the hardware in support of this. The algorithm is the recipe to treat the disease, the device is the instrument to administer it.”


A signal processing specialist designs the algorithm for a hearing aid from his office. He doesn't know the patient nor her hearing loss, her preferences or the common sound situations in her daily life. De Vries: “We have to deal with lots of uncertainties that cannot be resolved before patients actually wear their devices in the ‘real world’. The implication is that hearing aid algorithms carry a lot of tunable parameters in order to support these unknowns.” For many parameters, optimal values depend on the patient's (patho-)physiology and personal preferences. The hearing dispenser tunes these parameters together with his patient. But under the hood a hearing aid algorithm may contain about a hundred parameters. De Vries: “It is not helpful to expose a hearing aid practice to the complexity of fitting a hundred parameters in one session. The majority of parameters is therefore preset by the manufacturers. In practice, ten to fifteen most influential parameters are left to be tuned.” Tuning is usually done in a clinic or shop, not under real world acoustical conditions. De Vries: “There is no conclusive theory about how to tune parameters, there is no infallible fitting software nor a tool to make a perfect 'map' of someone's individual hearing loss. Except perhaps for the ‘compression ratio’ parameters, translating hearing measurements into device settings is therefore necessarily more an art based on professional feeling and experience than a scientific process.”

Machine learning

The overall conclusion, according to De Vries, is that a lot of uncertainty cannot be banned from the process. “Only the patient himself can hear what he hears. A hundred percent match of settings to the patient's real world situation is hard to attain. Patients sometimes search for years for the 'artist' who can find the right tuning for their situation. Over the past decade, more signal processing has not been able to change that, although the performance curve in algorithms has been steep thanks to the conventional 'feed forward' research approach. But the easy wins are over now, for further progress we have to change our methods.” The promise of machine learning is that the performance curve can be sustained, because it adds the patient's experience to the design process. Potential improvement of algorithms is locked inside the heads of hundreds of thousands of patients and could find its way via Big Data analysis methods into hearing aids. De Vries: “Some institutes already work with patients to learn from their experiences – dozens of them. But this can never equal the statistical significance we could attain with the far larger numbers brought within reach through wireless technology. Imagine a database with the hearing profiles of a hundred thousand volunteers: their audiogram and other auditory brain parameters, age, sex, musical preferences and measurements about their everyday sound environments.” These data could be uploaded for instance through smartphones. Device settings would also be entered in the database, followed by patient satisfaction scores after a year. De Vries: “That real world database would offer vital information for the bridge from measurements to parameter settings. It also opens the gate for a personalised approach. Which settings leads to what satisfaction rate for this or that category of patients? Machine learning is the field that extracts these patterns from such databases and the more data the better the analysis results get. It is what companies such as Amazon or Spotify already use when they offer book or song recommendations. Both the hearing aid industry and academia are currently setting up the necessary data science infrastructure for a likewise functionality.”

Higher satisfaction

Patients should have a free choice in helping to collect data. Eventually, also patients who are not willing to collaborate will benefit from better settings. “But we can offer the volunteers something extra”, says De Vries. “Let's say there is a hearing aid parameter that in principle can be set to any value between 1 and 10. The hearing aid dispenser or audiologist concludes that settings between 3 and 7 would support the audiological goals for his patient and he sends her away with the parameter set at 5. If she is not happy with her hearing aid, she can come back. In the new approach, we propose the following: the hearing aid specialist sets the parameter at 5, but enables his client to tweak the value by herself between 3 and 7 according to preference. The patient might eventually decide that 6 is the best setting. This limits the amount of visits to the shop, because the patient can choose the best setting by herself, which is convenient both for the specialist and client. We expect higher patient satisfaction scores as a result.” Moreover, this scenario enables the researchers to analyse the optimized settings and come up with better parameter settings, for instance a new adjustable range of the same parameter between 4 and 7, with initial setting at 6. De Vries: “Data analysis thus enables a cycle of continuous improvement on the basis of the combined effort by hearing aid manufacturer, audiologist and patient. All parties can benefit from this iterative learning loop.”

Changing roles

Nevertheless De Vries already expects that not every hearing aid professional will be overly happy with the prospect of such a development. “I understand that to some extent, because many roles are about to change as a result of the paradigm shift ahead of us. This change is comparable to the shift from analog to digital in the nineties. It came with hiccups, but after some years the progress became clear. With this change it will be like that. A hearing aid specialist may have doubts about his future position, as whiz kids with Big Data competence could shake up the hearing aid industry. But this is the road towards higher patient satisfaction rates, so if we don't go down the road of progress, others will for certain. Consumer electronics companies have also shown interest in the hearing care market. Manufacturers and specialists will have to make a switch if they want to play a role in the future hearing care market. Fortunately, paradigm shifts also come with many opportunities.” The hearing aid dispenser's position is not immediately at stake. De Vries: “It is a huge leap from the designer to the patient. If we cannot present adjustment of all possible settings to the professional, how could we expect the patient to do this entirely by himself? There will always be a need for professional expertise close to the patient. It will remain up to the professional to determine where his expertise will be most valuable in the forward and backward chains between manufacturer and patient.”
Who is Bert de Vries?

After his youth in Utrecht, Bert de Vries went to Eindhoven to study Electrical Engineering. He received an MSc degree in 1986 from Eindhoven University of Technology (TU/e) and a PhD in 1991 from the University of Florida in Gainesville, USA. During the nineties he worked at Sarnoff Research Center in Princeton, New Jersey, before returning to the Netherlands in 1999 to enter the hearing aid industry. He is currently a principal scientist at GN ReSound and a professor at TU/e.

Leendert van der Ent