Machine Learning for Vestibular Schwannoma Diagnosis Using Audiometrie Data Alone

Area of Science:

• Audiology
• Machine Learning
• Neuro-oncology

Background:

• Vestibular schwannomas are rare tumors affecting the auditory and balance nerves.
• Accurate diagnosis relies on interpreting audiometric data, often using established rule-based criteria.
• The efficacy of machine learning (ML) in this diagnostic context requires further investigation.

Purpose of the Study:

• To compare the diagnostic performance of ML algorithms against traditional rule-based evaluations for vestibular schwannoma screening using audiograms.
• To evaluate the performance of existing rule-based criteria in predicting vestibular schwannomas with the largest available dataset.

Main Methods:

• A retrospective case-control study was conducted at a tertiary referral center.
• Audiometric data from 767 adult patients with confirmed vestibular schwannoma and 2000 controls were analyzed.
• Data were processed using ML algorithms and standard rule-based criteria for asymmetric hearing loss.

Main Results:

• ML algorithms demonstrated mildly improved specificity in certain aspects compared to rule-based evaluations.
• Sensitivity of ML algorithms was comparable to previous findings for rule-based evaluations in diagnosing vestibular schwannomas.
• Rule-based formulas applied to the large dataset showed performance consistent with prior studies.

Conclusions:

• ML algorithms offer comparable diagnostic performance to rule-based evaluations for vestibular schwannoma detection based solely on audiometric data.
• Established rule-based formulas maintain their reliability and consistency when applied to extensive datasets.
• Both ML and rule-based approaches are valuable tools for audiogram-based screening of vestibular schwannomas.