Background: Early detection of cerebrovascular events (subarachnoid hemorrhage SAH, ischemic stroke, intracranial hemorrhage ICH) is critical for timely intervention. Emerging smartwatches with multiple sensors and artificial intelligence (AI) algorithms offer potential for real-time detection of acute neurological events. We performed a systematic review (2010–2025) per the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines to identify published studies on AI-enabled smartwatches for detecting SAH, ischemic stroke, and ICH. Methods: We searched multiple databases from January 2010 through May 2025 using combinations of smartwatch OR wearable , stroke OR SAH OR ICH , and AI OR machine learning OR deep learning . Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool for diagnostic accuracy studies. Results: The search yielded 3 eligible studies on ischemic stroke detection by wearable accelerometers with AI. The stroke studies (n = 3) used bilateral wrist/arm accelerometers to detect unilateral motor deficits. Deep learning models achieved high diagnostic accuracy (area under the receiver operating characteristic curve 0.95–0.99) for detecting acute stroke symptoms. One study reported a median detection time of 15 to 29 minutes after stroke onset, depending on the false alarm threshold. A feasibility trial (STROKE ALARM) using accelerometer bands and a smartphone app triggered frequent false alarms without observed strokes. Conclusion: Wearable technology with AI shows promise for ischemic stroke symptom detection, but there is a critical gap in SAH and ICH detection. Challenges include sensor accuracy, false alarms, and algorithm generalizability. We propose a conceptual multisensory model integrating heart rate (electrocardiogram/photoplethysmography), blood pressure surrogates, and motion data into an AI pipeline for future smartwatch systems, which can lead to the detection of stroke/SAH, might show promise in other brain pathologies.
Khan et al. (2026) studied this question.