Machine Learning Models Can Predict Tinnitus and Noise-Induced Hearing Loss.

Area of Science:

• Auditory science
• Medical artificial intelligence
• Diagnostic technology

Background:

• Machine learning (ML) applications in health sciences are widespread for prediction and classification.
• However, ML's use in differentiating auditory disorders is limited.
• This study addresses the gap in diagnosing tinnitus and distinguishing between noise-induced hearing loss (NIHL) and age-related hearing loss (ARHL).

Purpose of the Study:

• To evaluate the efficacy of five ML models in distinguishing tinnitus from non-tinnitus.
• To assess ML models' ability to differentiate NIHL from ARHL.
• To identify optimal ML models for auditory disorder diagnostics.

Main Methods:

• Utilized data from 928 Canadian adults (30-100 years) with diagnosed ARHL or NIHL.
• Applied five ML models: artificial neural networks (ANNs), K-nearest neighbors, logistic regression, random forest (RF), and support vector machines.
• Analyzed audiologic and demographic data, focusing on hearing loss patterns and tinnitus prevalence.

Main Results:

• Tinnitus prevalence was over double in the NIHL group (27.85% constant, 18.55% intermittent) compared to ARHL (8.85% constant, 10.86% intermittent).
• NIHL showed significantly greater hearing loss at medium- and high-band frequencies versus ARHL.
• ANN achieved 70% accuracy for tinnitus prediction; RF achieved 90% AUC for NIHL vs. ARHL differentiation.

Conclusions:

• ML models, particularly ANN and RF, enhance diagnostic precision for tinnitus and NIHL.
• Findings suggest a framework for integrating ML into clinical audiology.
• Future research should expand datasets and incorporate longitudinal data for broader applicability.