Gaussian Processes for Hearing Threshold Estimation Using Auditory Brainstem Responses.

Area of Science:

• Audiology
• Neuroscience
• Biomedical Engineering

Background:

• Auditory Brainstem Response (ABR) is crucial for hearing loss diagnosis but is time-consuming.
• Current ABR methods often fail to leverage waveform correlations across stimulus levels, limiting efficiency.
• Reducing ABR test duration is essential for clinical applications, especially for threshold estimation.

Purpose of the Study:

• To develop a more efficient method for estimating hearing thresholds using Auditory Brainstem Response (ABR).
• To exploit correlations in ABR waveforms across different stimulus levels for faster testing.
• To introduce a Bayesian approach, Gaussian Process (GP), for ABR analysis and hearing threshold estimation.

Main Methods:

• Utilized a Gaussian Process (GP) model for non-linear regression to estimate ABR amplitude across stimulus levels.
• Developed active learning rules to automatically adjust stimulus levels for efficient hearing threshold localization.
• Compared the GP method against a sequentially applied Hotelling's T^2 test in simulations and a case study.

Main Results:

• The GP approach demonstrated significant reductions in test time, up to approximately 50%, compared to the Hotelling's T^2 test.
• The method effectively utilizes correlations in ABR waveforms across stimulus levels for improved efficiency.
• A case study provided preliminary validation of the GP approach with human ABR data.

Conclusions:

• Gaussian Process modeling offers a promising and efficient approach for Auditory Brainstem Response (ABR) testing.
• This method can substantially reduce ABR test times, facilitating quicker and more effective hearing loss diagnosis.
• Exploiting waveform correlations across stimulus levels is key to optimizing ABR test efficiency.