Related Experiment Videos
Predicting hearing loss from otoacoustic emissions using an artificial neural network.
Rouviere de Waal1, René Hugo, Maggi Soer
1Department of Communication Pathology & Electronic Engineering, University of Pretoria.
Summary
Artificial neural networks (ANNs) accurately predict normal hearing using distortion product otoacoustic emissions (DPOAEs). This method shows promise for predicting hearing ability in individuals across a range of frequencies.
Area of Science:
- Audiology
- Biomedical Engineering
- Machine Learning
Background:
- Distortion product otoacoustic emissions (DPOAEs) are objective measures of cochlear function.
- Pure tone thresholds (PTTs) are standard measures of hearing sensitivity.
- Artificial neural networks (ANNs) offer advanced pattern recognition capabilities.
Purpose of the Study:
- To predict pure tone thresholds (PTTs) from distortion product otoacoustic emissions (DPOAEs) using an artificial neural network (ANN).
- To evaluate the accuracy of ANN-based PTT prediction for normal and impaired hearing across various frequencies.
Main Methods:
- An artificial neural network (ANN) with back-propagation was trained using DPOAE data from eight DPgrams (406-4031 Hz).
- The ANN predicted PTTs at 0.5, 1, 2, and 4 kHz based on the presence or absence of DPOAEs.
- Normal hearing was defined as < 25 dB HL.
Main Results:
- Prediction accuracy for normal hearing was high: 94% at 500 Hz, 88% at 1000 Hz, 88% at 2000 Hz, and 93% at 4000 Hz.
- Prediction accuracy for hearing-impaired categories was lower due to limited training data.
- The ANN demonstrated the potential to predict hearing ability within 10 dB.
Conclusions:
- ANNs coupled with DPOAEs can accurately predict hearing ability in normal-hearing individuals.
- This approach shows potential for predicting hearing loss in impaired listeners.
- The study highlights the feasibility of using DPOAEs and ANNs for hearing assessment from 500-4000 Hz.