A machine learning model based on pure-tone audiometry features can diagnose Meniere disease (MD) and predict endolymphatic hydrops (EH), according to a study published online Aug. 28 in Otolaryngology-Head and Neck Surgery . Xu Liu, M.
D., from Fudan University in Shanghai, and colleagues collected gadolinium-enhanced magnetic resonance imaging sequences and pure-tone audiometry data in a retrospective study . Based on the air conduction thresholds of pure-tone audiometry, basic and multiple analytical features were engineered.
The engineered features were used to train five classical machine learning models to diagnose MD. The models with excellent performance were selected to predict EH. The researchers found that the winning light gradient boosting (LGB) machine learning model demonstrated remarkable performance for diagnosis of MD, achieving accuracy of 87%, sensitivity and specificity of 83 and 90%, and an area under the receiver operating characteristic curve of 0.
95, comparing favorably with experienced clinicians. The LGB model had 78% accuracy for EH prediction and outperformed the other three models. Specific pure-tone audiometry features that are essential for both MD diagnosis and EH prediction include standard deviation and mean of the whole-frequency hearing, audiogram peak, and hearing at low frequencies (notably at 250 Hz).
"The study showed promising diagnostic capabilities, indicating the potential of machine learning based on pure‐tone audiometry results .
