Machine learning based pure-tone audiometry effective in diagnosing Meniere's disease: Study

A new study published in the journal of Otolaryngology-Head and Neck Surgery showed that an effective machine learning model utilizing pure-tone audiometry characteristics effectively diagnosed Meniere's disease (MD) with the ability to foresee the subtypes of endolymphatic hydrops.

One of the most complicated disorders in otolaryngology is Meniere's disease which is a complex vestibular dysfunction that presents significant diagnostic hurdles. As of now, the etiology of MD is complex and poorly understood with endolymphatic hydrops (EH) as a key pathohistological characteristic. Even though EH may presently be seen using gadolinium-enhanced MRI, clinical symptoms still play a major role in the diagnosis of MD.

When MD is first diagnosed, it is frequently divided into "definite MD" and "probable MD." Thus, Xu Liu and team conducted this study with the purpose to deploy machine learning models for the automated diagnosis of Meniere's illness and the prediction of endolymphatic hydrops, based on the air conduction thresholds of pure-tone audiometry.

Pure-tone audiometry data and gadolinium-enhanced magnetic resonance imaging sequences were gathered. Based on the air conduction thresholds of pure-tone audiometry, basic and various analytical characteristics were then constructed. Afterwards, the constructed characteristics were used to train 5 traditional machine learning models for MD diagnosis. The models that performed exceptionally well were also chosen to forecast EH. The performance of the model in diagnosing MDs was evaluated against that of skilled otolaryngologists.

With an accuracy rate of 87%, sensitivity of 83%, specificity of 90%, and a strong area under the receiver operating characteristic curve of 0.95, the winning light gradient boosting (LGB) machine learning model trained by multiple features performs remarkably well on the diagnosis of MD.

Also, the LGB model was better than the other 3 machine learning models, with an accuracy of 78% on EH prediction. Further, a feature significance analysis highlights the critical relevance of certain pure-tone audiometry parameters that are necessary for EH prediction and MD diagnosis. At 250 Hz, the standard deviation and mean of the whole-frequency hearing, hearing at low frequencies and the peak audiogram stood out. Overall, this study supports the use of pure-tone audiometry data in an AI-based method to distinguish between MD. 

Reference:

Liu, X., Guo, P., Wang, D., Hsieh, Y., Shi, S., Dai, Z., Wang, D., Li, H., & Wang, W. (2024). Applications of Machine Learning in Meniere’s Disease Assessment Based on Pure‐Tone Audiometry. In Otolaryngology–Head and Neck Surgery. Wiley. https://doi.org/10.1002/ohn.956