Periodic Reporting for period 4 - IBiDT (Individualized Binaural Diagnostics and Technology)
Reporting period: 2022-07-01 to 2023-12-31
360 million people worldwide have some form of hearing loss, 10% of them children. While hearing aids and cochlear implants reasonably restore the hearing with the ear they supply, they do often not restore two things that normal hearing listeners take for granted: 1) Being able to localize a sound source in space and 2) being able to understand a conversation partner even in the presence of more intense competing talkers and other sound sources. Both of these abilities have been shown to contribute to the quality of life.
A sound arriving from a certain direction in space (e.g. off the midline) arrives at the two ears at slightly different times, generating an interaural time difference (ITD). In addition, the head attenuates the sound at the ear opposite to the sound source to produce an interaural level difference (ILD). Brainstem circuits are specialized in detecting ITDs and ILDs with a temporal precision unique within the nervous system. As soon as one or two hearing aids or cochlear implants are applied to the ears, any difference in their sound processing corrupts the interaural differences, reducing the binaural advantage. Further, a fair number of people that are said to be “normal hearing” in clinical terms does not get the “normal” advantage of listening with two ears – binaural hearing. This includes particularly people with a neurologic history, such as a stroke but can also happen in completely healthy individuals. To date hearing researchers can list many factors that may limit the binaural benefit. However, due to the complexity of the auditory system (a complex inner ear followed by an even more complex brain), they have only a limited understanding in which way and in which situation a certain pathology corrupts binaural hearing. Hearing devices are far from employing such causalities for optimizing algorithms to the specific pathology.
The objective of the project is therefore to better relate neurophysiology and pathophysiology to normal and impaired perception. This is done by developing a detailed computer simulation of the auditory system validated through a broad range of listening experiments. Once we are confident that we know the pathology of an individual we can then simulate this impairment and see when and how it causes the most extreme and maybe a unique problem. More though, we can develop algorithms, similar to those in hearing aids, but specific to the pathology, optimize them with the simulation and test them with the patient. If the patient shows the same improvement we have predicted in our simulation, we have not only helped the patient but demonstrated that our novel approach can potentially help hundreds of thousand people to improve their binaural hearing and hence their quality of life. Our team is thrilled to pioneer this approach and hope to trigger many follow-up studies.
A sound arriving from a certain direction in space (e.g. off the midline) arrives at the two ears at slightly different times, generating an interaural time difference (ITD). In addition, the head attenuates the sound at the ear opposite to the sound source to produce an interaural level difference (ILD). Brainstem circuits are specialized in detecting ITDs and ILDs with a temporal precision unique within the nervous system. As soon as one or two hearing aids or cochlear implants are applied to the ears, any difference in their sound processing corrupts the interaural differences, reducing the binaural advantage. Further, a fair number of people that are said to be “normal hearing” in clinical terms does not get the “normal” advantage of listening with two ears – binaural hearing. This includes particularly people with a neurologic history, such as a stroke but can also happen in completely healthy individuals. To date hearing researchers can list many factors that may limit the binaural benefit. However, due to the complexity of the auditory system (a complex inner ear followed by an even more complex brain), they have only a limited understanding in which way and in which situation a certain pathology corrupts binaural hearing. Hearing devices are far from employing such causalities for optimizing algorithms to the specific pathology.
The objective of the project is therefore to better relate neurophysiology and pathophysiology to normal and impaired perception. This is done by developing a detailed computer simulation of the auditory system validated through a broad range of listening experiments. Once we are confident that we know the pathology of an individual we can then simulate this impairment and see when and how it causes the most extreme and maybe a unique problem. More though, we can develop algorithms, similar to those in hearing aids, but specific to the pathology, optimize them with the simulation and test them with the patient. If the patient shows the same improvement we have predicted in our simulation, we have not only helped the patient but demonstrated that our novel approach can potentially help hundreds of thousand people to improve their binaural hearing and hence their quality of life. Our team is thrilled to pioneer this approach and hope to trigger many follow-up studies.
We have deloped a comprehensive model of binaural processing (Klug et al. 2020, Encke and Dietz, 2022, Eurich et al. 2023) and data to support it (Dietz et al. 2021, Klug and Dietz, 2023). The concept was also translated to model binaural hearing with two cochlear implants (Hu et al. 2022).
We have developed the concept of "model-based action selection" (Herrmann and Dietz (2021). The model is update "on the fly" during experimental data collection. The model then tells wich of its parameters describing the current patient are still uncertain and which experiment needs to be done next to have the best expected reduction of uncertainty. It has been tested in normal hearing subjects (Dietze et al. 2024).
We published a comprehensive open-science framework for simulating Cochlear Implants and modelling electric hearing. We took the lead in an international collaboration with researchers from Cambridge, UK and Miami, FL, USA. (Hu et al. 2023, Acta Acustica).
We conducted a first of its kind large scale longitudinal study on spatial hearing deficits after stroke (Dietze et al. (2022+2024) Frontiers in Neuroscience.)
Overall, the project has already led to 17 peer-reviewed publications, 13 of which are rooted in the project, i.e. they started during the project, the first author was employed on the project and I was the last author (once first author). More manuscripts are at different stages in the pipeline.
We have developed the concept of "model-based action selection" (Herrmann and Dietz (2021). The model is update "on the fly" during experimental data collection. The model then tells wich of its parameters describing the current patient are still uncertain and which experiment needs to be done next to have the best expected reduction of uncertainty. It has been tested in normal hearing subjects (Dietze et al. 2024).
We published a comprehensive open-science framework for simulating Cochlear Implants and modelling electric hearing. We took the lead in an international collaboration with researchers from Cambridge, UK and Miami, FL, USA. (Hu et al. 2023, Acta Acustica).
We conducted a first of its kind large scale longitudinal study on spatial hearing deficits after stroke (Dietze et al. (2022+2024) Frontiers in Neuroscience.)
Overall, the project has already led to 17 peer-reviewed publications, 13 of which are rooted in the project, i.e. they started during the project, the first author was employed on the project and I was the last author (once first author). More manuscripts are at different stages in the pipeline.
Two main breakthroughs have been achieved:
1) My field was devided into two camps because physiologists had evidence that the standard model of binaural processing (Jeffress 1948) is wrong but many psychologists and engineers remained in favor of the model because it could explain most behavioral data. Our new model (Klug et al. 2020, Encke and Dietz, 2022) appears to re-unite the field because it is built on the physiologic findings and accounts for behavioral data even better than the controversial former standard model. It is also computationally efficient which is what the engineers were looking for. I have been working on this problem since my own PhD thesis research, starting 2006, so it was very planned but only because the ERC grant allowed me to attract a talented postdoc (who worked on the same issue in his PhD thesis), we we finally able to solve the problem.
2) When writing the proposal, the concept for model-based patient performance prediction was at the very core of the proposal, connecting 5 of 6 work packages. When putting it into action we realized that the two measurement stages (1) untargeted measurements - the same for all and (2) targeted measurements to parametrize the indivual impairment are only a simplification of what is actually ideal: Make the measurements gradually more and more targeted with a model running in the background while data is collected. Our model-based action selection has the potential to revolutionize auditory experimentation as well as audiologic diagnostics. We expect that the new method will result in improving time efficiency by a factor of two, if not more.
1) My field was devided into two camps because physiologists had evidence that the standard model of binaural processing (Jeffress 1948) is wrong but many psychologists and engineers remained in favor of the model because it could explain most behavioral data. Our new model (Klug et al. 2020, Encke and Dietz, 2022) appears to re-unite the field because it is built on the physiologic findings and accounts for behavioral data even better than the controversial former standard model. It is also computationally efficient which is what the engineers were looking for. I have been working on this problem since my own PhD thesis research, starting 2006, so it was very planned but only because the ERC grant allowed me to attract a talented postdoc (who worked on the same issue in his PhD thesis), we we finally able to solve the problem.
2) When writing the proposal, the concept for model-based patient performance prediction was at the very core of the proposal, connecting 5 of 6 work packages. When putting it into action we realized that the two measurement stages (1) untargeted measurements - the same for all and (2) targeted measurements to parametrize the indivual impairment are only a simplification of what is actually ideal: Make the measurements gradually more and more targeted with a model running in the background while data is collected. Our model-based action selection has the potential to revolutionize auditory experimentation as well as audiologic diagnostics. We expect that the new method will result in improving time efficiency by a factor of two, if not more.