A machine-learning-based approach to predict early hallmarks of progressive hearing loss

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Genetics

Background:

• Age-related hearing loss (ARHL) is a prevalent sensory deficit in adults, yet early diagnosis remains challenging.
• Current diagnostic methods for ARHL may not detect subtle, early-stage changes.
• Machine learning (ML) offers potential for analyzing complex biological data to identify early disease markers.

Purpose of the Study:

• To investigate the utility of ML for the early detection of ARHL in a mouse model.
• To evaluate ML models' ability to distinguish between mice with and without genetic predisposition to ARHL.
• To explore ML's capacity for predicting the rate of ARHL progression.

Main Methods:

• Auditory brainstem responses (ABRs) were recorded from C57BL/6N mice with early-onset ARHL (Cdh23^ahl allele) and control mice (Cdh23+).
• Various ML classifiers were trained and evaluated to differentiate between the two genotypes based on ABR data.
• Shapley values were used for feature importance analysis to identify key ABR characteristics for classification.

Main Results:

• ML models accurately identified mice with the Cdh23^ahl allele at 1 month of age, prior to overt hearing loss.
• Machine learning classification performance surpassed that of human experts.
• Feature importance analysis highlighted subtle ABR wave 1 differences as critical discriminators.
• Regression models successfully predicted future ARHL progression rates from early ABR recordings.

Conclusions:

• ML techniques demonstrate significant potential for the early diagnosis of ARHL by detecting subtle auditory changes.
• ML can identify individuals at risk for ARHL and predict disease progression.
• These findings suggest ML could enhance the efficacy of future ARHL treatments in humans.