Machine Learning Based Diagnosis of Vertigo using Video Head Impulse Test

This work utilises the strength of state space based dynamic modelling and the ability of machine learning based segmentation of SRM standard descriptors to reach superior diagnostic capabilities. Dynamic modelling ensured vHIT input-output characteristics generated SRM standard descriptors, which were consequently used in formation of ML classification models.The best ML model was Linear SVM when built on left and right sided data with the SRM standard descriptors: rise time, settling time, settling minimum, settling maximum, overshoot and undershoot. The model was able to classify individuals to patient or control groups with an accuracy of 100% and a sensitivity and specificity of 1.Clinical Relevance- Dizziness is one of the most common presentations to family physicians and emergency departments. It is associated with significant medical complications such as falls as well as economic costs to both the individual and the community. Vestibular diseases comprise the bulk of dizzy disorders and are often associated with dysfunction of the vestibular or inner ear balance apparatus. This is most commonly the result of hypo-function of the semi-circular canals. Clinically, the most commonly employed objective test of semicircular function is the video Head Impulse Test (vHIT). Here we provide a machine learning approach to a more comprehensible and accurate interpretation of the results obtained by the vHIT to more robustly establish the presence and severity of VOR dysfunction, and ultimately aid in the diagnosis of vestibular disorders.