Prediction of Vestibular Dysfunction by Applying Machine Learning Algorithms to Postural Instability.

Area of Science:

• Neuroscience and Biomedical Engineering
• Vestibular System Disorders
• Machine Learning in Healthcare

Background:

• Peripheral vestibular dysfunction is a common cause of dizziness.
• Accurate diagnosis of vestibular dysfunction is crucial for effective treatment.
• Traditional diagnostic methods can be time-consuming and may require specialized equipment.

Purpose of the Study:

• To evaluate the efficacy of various machine learning algorithms in predicting peripheral vestibular dysfunction.
• To utilize center of pressure (COP) sway data from foam posturography for prediction.
• To compare the performance of Gradient Boosting Decision Tree, Bagging Classifier, and Logistic Regression.

Main Methods:

• Retrospective study involving 75 patients with vestibular dysfunction and 163 healthy controls.
• COP sway data (velocity, envelopment area, power spectrum) analyzed using K-fold cross-validation.
• Algorithms trained and validated using Area Under the Curve (AUC) and recall metrics.

Main Results:

• Gradient Boosting Decision Tree achieved the highest AUC (0.90 ± 0.06) and recall (0.84 ± 0.07).
• Both AUC and recall for Gradient Boosting Decision Tree were significantly higher than logistic regression.
• Bagging Classifier also demonstrated significantly higher recall (0.82 ± 0.07) compared to logistic regression.

Conclusions:

• Machine learning algorithms, particularly Gradient Boosting Decision Tree, can effectively predict peripheral vestibular dysfunction.
• COP sway data from posturography is a valuable dataset for machine learning-based diagnostic tools.
• Algorithm performance varies; evaluation and hyperparameter optimization are essential for clinical application.