内耳水腫の診断におけるマルチモーダル聴覚データを用いた解釈可能な機械学習
Xu Liu1,2, Qin Sun1,2, Suming Shi1,2
1ENT Institute, Department of Otorhinolaryngology, Eye & ENT Hospital, Fudan University, Shanghai, 200031, China.
Objective:
This study aimed to assess the diagnostic value of electrocochleography (ECochG) for identifying endolymphatic hydrops (EH) and to investigate whether integrating multimodal auditory data from ECochG and pure-tone audiometry (PTA) into machine learning models could improve diagnostic accuracy and clinical interpretability.
Methods:
A prospective cohort of 78 patients (156 ears) with a strong clinical suspicion of EH was evaluated between March and June 2024. ECochG and PTA examinations were performed, and extracted parameters were used to develop and validate multiple machine learning models. Diagnostic performance was compared using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and overall accuracy.
Results:
After excluding six ears with unrecordable ECochG data, 150 ears were analyzed. Models based solely on ECochG-derived features demonstrated limited diagnostic performance (maximum AUC = 0.61). When multimodal hearing data were incorporated, model performance improved substantially. The Light Gradient Boosting Machine (LightGBM) model achieved an AUC of 0.92 and an accuracy of 0.73 for predicting the anatomical distribution of EH.
Conclusions:
Integrating multimodal auditory data into interpretable machine learning models can markedly improve the diagnostic accuracy of endolymphatic hydrops, providing an objective framework to support early clinical diagnosis and management. However, this study is limited by a small sample size and has not yet been validated across multiple centers.
Level Of Evidence:
Level 3.
