Performance and reliability evaluation of an improved machine learning-based pure-tone audiometry with automated
Nicolas Wallaert1,2, Antoine Perry2,3, Sandra Quarino1
1Department of Otorhinolaryngology-Head and Neck Surgery Rennes University Hospital Rennes France.
Summary
This study introduces an improved machine learning (ML) audiometry method for air-conduction pure-tone audiograms. The enhanced ML approach demonstrates comparable accuracy to manual audiometry across various hearing statuses, validating its clinical use.
Area of Science:
- Audiology
- Machine Learning
- Signal Processing
Background:
- Automated audiometry using machine learning (ML) shows promise but has limitations for specific hearing conditions.
- Existing ML approaches are unsuitable for patients with asymmetric, conductive, or severe hearing loss, or cochlear dead zones.
Purpose of the Study:
- To enhance ML-based audiometry for broader clinical applicability.
- To evaluate the performance of an improved ML audiometry method in a large, diverse listener cohort.
Main Methods:
- Modified ML algorithm incorporating safety limits, transient responses, and automated contralateral masking.
- Comparison of the improved ML audiometry against conventional and manual audiometry in 109 participants.
Main Results:
- No significant difference in hearing thresholds between manual and improved ML methods across tested frequencies.
- The automated improved ML method showed no significant test-retest variability.
- Cross-clinic reliability revealed significant differences at most audiometric frequencies.
Conclusions:
- The improved ML-based method is validated for adult clinical air-conduction audiometry.
- This enhanced ML approach offers a reliable alternative to traditional audiometric methods.
