Machine learning models using combinations of history, examination, and vestibular tests can accurately separate posterior circulation stroke and acute unilateral vestibulopathy, achieving accuracies of up to 96.6% and holding promise as diagnostic aids for frontline clinicians.
The best-performing machine learning models identified posterior circulation stroke with high accuracy across all three clinical tiers.
HINTS testing by experts achieved 94.6% accuracy in differentiating posterior circulation stroke from acute unilateral vestibulopathy.
The study recruited 294 patients presenting to the Emergency Room with acute vestibular syndrome.
The most important predictive variables differed across the three clinical tiers.
Three hierarchical clinical scenarios (tiers) were defined to simulate different levels of clinical expertise and available resources.
Machine learning models were developed and evaluated using a broad array of data inputs spanning history, bedside examination, and specialized vestibular testing.
This research suggests that machine learning—a type of artificial intelligence—can accurately distinguish between two common causes of sudden, severe dizziness: a benign inner ear condition called acute unilateral vestibulopathy (AUVP) and a dangerous posterior circulation stroke (PCS). Both conditions can look very similar when patients arrive in the emergency room, making correct diagnosis difficult and time-sensitive. The study enrolled 294 patients at an emergency room and used their medical history, physical examination findings, and specialized balance and hearing tests as inputs to train and test several AI models. The AI models performed very well across different hospital settings. In hospitals with the most resources and specialist support, the model was correct 96.6% of the time. Even in emergency rooms with only basic history-taking and physical examination—no specialized equipment—the model still achieved 88.8% accuracy. These results were comparable to or better than the accuracy achieved by expert clinicians using a standard bedside test called HINTS (94.6%). The study also found that the most useful pieces of information for making the diagnosis varied by setting: in well-equipped centers, the bedside head-impulse test and presence of neurological symptoms were most important, while in lower-resource settings, patient age played the biggest role. This research suggests that AI-based tools could serve as practical decision-support aids for frontline emergency clinicians—particularly those without specialized neurology or ear-nose-throat expertise—helping them more accurately identify which dizzy patients are having a stroke and need urgent treatment. The tiered approach is especially promising because it acknowledges that not all emergency rooms have access to the same technology or specialists, and shows the AI can be useful even with limited information.
Wang C, Chaturvedi K, Nham B, Reid N, Bradshaw A, Rosengren S, et al.. (2026). Separating stroke and acute unilateral vestibulopathy using history, examination and vestibular tests: a machine learning approach.. Journal of neurology. https://doi.org/10.1007/s00415-026-13779-0