2025 American Academy of Neurology Abstract Website

Chao Wang¹, Jeevan Sreerama¹, Benjamin Nham³, Nicole Reid⁵, Nese Ozalp⁶, James Thomas⁶, Cecilia Cappelen-Smith⁴, Zeljka Calic⁴, Andrew Bradshaw⁵, Sally Rosengren¹, Gulden Akdal⁷, Michael Halmagyi¹, Deborah Black², David Burke¹, Mukesh Prasad⁸, Gnana Bharathy⁸, Miriam Welgampola¹
¹Central Clinical School, ²Faculty of Medicine and Health, University of Sydney, ³St George and Sutherland Clinical School, ⁴South Western Sydney Clinical School, University of New South Wales, ⁵Institute of Clinical Neurosciences, Royal Prince Alfred Hospital, ⁶Department of Neurophysiology, Liverpool Hospital, ⁷Department of Neurology, Dokuz Eylül University, ⁸School of Computer Science, University of Technology Sydney

Objective:

To develop and evaluate machine learning models that can differentiate between posterior-circulation stroke and vestibular neuritis using only the video head impulse test (VHIT).

Background:

Acute Vestibular Syndrome, a sudden severe episode of vertigo and imbalance lasting 24 hours or longer, usually represents Vestibular Neuritis (VN), an innocuous viral illness, or Posterior-Circulation Stroke (PCS), a potentially life-threatening event. The VHIT is a quantitative measure of the vestibulo-ocular reflex that can distinguish between these two diagnoses. It can be rapidly performed by any trained healthcare professional at the bedside but requires interpretation by an expert clinician. Machine learning models that can use VHIT to separate VN and PCS with expert-level accuracy would help frontline clinicians without access to neuro-otology expertise.

Design/Methods:

We trained machine learning classification models using unedited (raw) head and eye-velocity traces from acute VHIT performed in an Emergency Room on patients presenting with acute vestibular syndrome and whose final diagnosis was vestibular neuritis or posterior-circulation stroke. The models were validated using an independent test dataset collected at a second institution. We also compared the performance of machine learning models against expert clinicians as well as a widely used VHIT metric: the gain cut-off value.

Results:

The training and test datasets comprised 257 and 49 patients respectively. In the test dataset, the best machine learning model identified vestibular neuritis with 87.8% (95% CI 77.6%-95.9%) accuracy. Model performance was not significantly different (p=0.56) from that of 4 blinded expert clinicians who achieved 85.7% accuracy (75.5%-93.9%) and was superior (p=0.01) to that of the optimal gain cut-off value (75.5% accuracy (63.8%-85.7%)).

Conclusions:

Machine learning models can effectively differentiate posterior-circulation stroke from vestibular neuritis using only VHIT data, with comparable accuracy to expert clinicians. They hold promise as an Emergency Room tool which could assist frontline clinicians evaluating patients with acute vestibular syndrome.