Did You Hear That? Detecting Auditory Events with EEGNet.

Area of Science:

• Neuroscience
• Machine Learning
• Audiology

Background:

• Pure-tone audiometry (PTA) relies on behavioral responses, limiting its use in individuals with physical or cognitive disabilities.
• Existing brain signal analysis methods, like Event Related Potentials (ERPs), have shown limited utility for identifying hearing thresholds.
• Accurate hearing threshold measurement is crucial for diagnosing and managing hearing loss across diverse patient populations.

Purpose of the Study:

• To investigate the efficacy of EEGNet, a convolutional neural network, in detecting auditory events from electroencephalography (EEG) data.
• To compare EEGNet's performance against traditional machine learning models like Support Vector Machines (SVMs) and Common Spatial Patterns + Linear Discriminant Analysis (CSPLDA).
• To assess the potential of a brain-signal-based hearing test for individuals unable to participate in conventional audiometric assessments.

Main Methods:

• EEG data was collected from participants undergoing a simulated pure-tone audiogram test.
• EEGNet, SVMs, and CSPLDA models were trained on the collected EEG dataset for hearing event detection.
• Model performance was evaluated on unseen participants, comparing accuracy and statistical significance.

Main Results:

• EEGNet achieved 81.5% accuracy in detecting hearing events in unseen participants, outperforming SVMs by over 5%.
• While EEGNet showed superior performance, the improvement over SVMs and CSPLDA was not always statistically significant.
• Further analysis indicated EEGNet's potential for accurate hearing threshold determination with sufficient test repetitions.

Conclusions:

• EEGNet demonstrates promise as a tool for developing a brain-signal-based hearing test, expanding audiological assessment capabilities.
• This approach could significantly benefit individuals with disabilities that impede participation in standard behavioral hearing tests.
• Future research is necessary to optimize test setup, reduce testing duration, and further enhance detection accuracy.