Document resource
Background/Objectives Acute Vestibular Syndrome usually represents either Vestibular Neuritis (VN) or Posterior Circulation Stroke (PCS). The video head impulse test (VHIT) is a quantitative test of the vestibulo-ocular reflex that can distinguish between these 2 diagnoses. It can be rapidly performed at the bedside by any trained healthcare professional but requires interpretation by an expert clinician. We developed machine learning models to differentiate PCS from VN using only the VHIT.Methods Machine learning classification models were trained using unedited (raw) head- and eye-velocity data from acute VHIT performed on Emergency Department patients who presented with acute vestibular syndrome and were diagnosed as VN or PCS. The models were validated using an independent test dataset from a second institution. We compared the performance of the models against both expert clinicians as well as a widely used VHIT metric: the gain cut-off value.Results The training and test datasets included 257 and 49 patients respectively. The best machine learning model used the ROCKET algorithm and identified VN with 87.8% accuracy (95% CI: 77.6%-95.9%) in the test dataset. Its performance was not significantly different (p=0.56) from blinded expert clinicians using VHIT (85.7% accuracy, 95% CI: 75.5%-93.9%) and was superior (p=0.01) to that of the optimal gain cut-off value (75.5% accuracy, 95% CI: 63.8%-85.7%).Conclusion Using only VHIT data, machine learning models can separate VN from PCS with comparable accuracy to experts. They hold promise as diagnostic aids for non-expert Emergency Department clinicians evaluating patients with acute vestibular syndrome.