Multidimensional Feature Analysis of Meniere's Disease and Vestibular Migraine: Insights from Machine Learning and Vestibular Testing.

Objective: Differentiating between Meniere's disease (MD) and vestibular migraine (VM) is challenging due to overlapping symptoms and limited diagnostic tools. Traditional statistical methods often rely on physician judgment and struggle with complex, high-dimensional data. This study applies the random forest (RF) machine learning algorithm to enhance the clinical differentiation between MD and VM.

Methods: We retrospectively analyzed data from 36 VM (26 female) and 100 unilateral MD patients (51 female). The data were anonymized and labeled. Symptomatic and examination parameters were selected as features, and exploratory data analysis identified key parameters for diagnosis. An RF model was used to rank these features.

Results: MD patients more commonly experienced ear-related symptoms, while VM patients reported more headaches and dizziness. Examination findings showed greater asymmetry in vHIT saccade latency in MD patients, particularly on the affected side. A total of 40 key parameters were identified. Heatmap and clustering analysis revealed that time constant (Tc) in velocity step test (VST) correlated more strongly with headache and other symptoms, while saccade latencies and velocities correlated with pure tone averages. The RF model selected 27 parameters for prediction, achieving 91.86% accuracy (95% confidence interval [85.37%, 95.18%]). Tc and saccade velocity were among the top 10 contributing features. Additionally, MD patients had earlier saccades and shorter Tc values on the affected side compared to both healthy controls and VM patients.

Conclusions: Machine learning successfully classified MD and VM patients, with Tc and saccade velocity identified as key diagnostic indicators alongside symptoms.