Automatic Recognition of Auditory Brainstem Response Waveforms Using a Deep Learning-Based Framework

Area of Science:

• Audiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Auditory Brainstem Response (ABR) waveform recognition is crucial for audiological assessment but can be challenging for certain populations.
• Deep learning offers potential solutions for improving the accuracy and efficiency of ABR analysis.

Purpose of the Study:

• To investigate the effectiveness of deep learning frameworks for automatic ABR waveform recognition.
• To assess performance across diverse age groups and hearing loss levels.

Main Methods:

• A descriptive study analyzed pure tone audiometry and ABR data from 100 participants.
• Feature vectors were generated using time-domain, frequency-domain ABR signals, and demographic data.
• An enhanced Wide&Deep model incorporating a Light-multi-layer perceptron (MLP) was used for ABR waveform recognition.

Main Results:

• The Light-MLP model achieved a weighted average accuracy of 95.4%.
• The Wide&Deep model achieved a weighted average accuracy of 91.0%.
• Both models demonstrated high recognition accuracy across different participant groups.

Conclusions:

• Deep learning models, specifically Light-MLP and Wide&Deep, show significant promise for accurate ABR waveform recognition.
• The Light-MLP model outperformed the Wide&Deep model in this study.
• Further data augmentation may enhance Wide&Deep model performance.