Machine Learning Based Diagnosis of Vertigo using Video Head Impulse Test

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Dizziness is a common symptom, often caused by vestibular disorders affecting the inner ear's balance function.
• The video Head Impulse Test (vHIT) is a key clinical tool for assessing semicircular canal function.
• Accurate interpretation of vHIT results is crucial for diagnosing Vestibulopathy and dysfunction of the vestibulo-ocular reflex (VOR).

Purpose of the Study:

• To develop a machine learning (ML) approach for enhanced interpretation of vHIT data.
• To improve diagnostic accuracy for vestibular disorders using dynamic modeling and ML.
• To establish a robust method for assessing VOR dysfunction severity.

Main Methods:

• Utilized state-space dynamic modeling to generate vHIT input-output characteristics.
• Employed machine learning-based segmentation of Standardized Rotary Motion (SRM) descriptors.
• Trained Linear Support Vector Machine (SVM) models using SRM descriptors like rise time, settling time, overshoot, and undershoot.

Main Results:

• The Linear SVM model, using specific SRM descriptors, achieved 100% accuracy in classifying individuals into patient or control groups.
• The model demonstrated perfect sensitivity and specificity (1.0) for classification.
• This approach offers superior diagnostic capabilities for vestibular disorders.

Conclusions:

• Machine learning analysis of vHIT data significantly enhances diagnostic accuracy for vestibular dysfunction.
• The developed model provides a more comprehensible and accurate interpretation of vHIT results.
• This method aids in the robust diagnosis and assessment of VOR dysfunction, benefiting clinical practice for dizzy patients.