On the feasibility of using neural nets to derive hearing-aid prescriptive procedures

Area of Science:

• Audiology
• Artificial Intelligence
• Signal Processing

Background:

• Neural networks offer a powerful approach for complex data analysis.
• Hearing-aid fitting requires precise matching of device characteristics to individual hearing loss.
• Traditional fitting methods may not always achieve optimal outcomes for all hearing loss profiles.

Purpose of the Study:

• To evaluate the feasibility of using neural networks for hearing-aid fitting.
• To train a multilayer perceptron net to determine optimal hearing-aid frequency response and gain.
• To compare the accuracy of neural network-based fitting with a standard procedure (NAL-R).

Main Methods:

• A multilayer perceptron neural network was trained using pure-tone audiogram data.
• Simulated and real audiometric data were used to test the neural network's performance.
• The neural network's predictions were compared against the NAL-R fitting procedure.

Main Results:

• Neural networks can be trained to replicate established hearing-aid fitting rules like NAL-R.
• Approximately 50 datasets are needed for the neural network to achieve a generalized solution.
• The neural network demonstrated higher accuracy than the NAL-R procedure in predicting optimal characteristics for severe-to-profound hearing losses.

Conclusions:

• Neural networks show significant potential for improving hearing-aid fitting accuracy.
• This AI-driven approach can provide more personalized and effective hearing solutions.
• Further research can explore broader applications of neural networks in audiology.