Artifacts Extraction From Video Head Impulse Test Data Using Time Series Classification Methods and VOR Gain Analysis

The video head impulse test (vHIT) has become an essential tool in the examination of patients with dizziness and other balance disorders, providing significant data on all six semicircular canals. The clinical interpretation of the vestibulo-ocular reflex (VOR) dynamic function of the human brain in vertigo and balance disorders using the vHIT method poses a considerable challenge. We utilize VOR gain measurements to ascertain the health of the patient’s vestibular system. However, all methods have inherent limitations due to the presence of noise and artifacts in the data, which can significantly affect the gain values of normal and abnormal impulses, leading to inaccuracies. This paper presents a comprehensive study, where we have created a dataset using vHIT data from 5,782 clinical patients from the Department of Otorhinolaryngology, College of Medicine, Inha University. We apply time series classification (TSC) algorithms to identify and filter artifact-affected impulses, ensuring more reliable VOR gain calculations. The encoder model achieved a classification accuracy of 94%, surpassing previous approaches such as SSNHLV (92%) and AI-based stroke (88%) classification. Statistical analysis confirms the significance of our method, with p-values (<0.05) demonstrating a clear distinction between normal, abnormal, and artifact impulses. By improving impulse classification, our approach enhances the precision of VOR gain calculations, contributing to more accurate clinical diagnoses of vestibular disorders.