Artificial Intelligence‐Based Differentiation of Ischemic and Hemorrhagic Stroke: A Retrospective, Analytical and Observational Case‐Control Design

Background: Rapid differentiation between ischemic and hemorrhagic stroke is essential because management differs significantly. In settings without immediate neuroimaging, artificial intelligence (AI) may assist early decision‐making. To evaluate the performance of AI in distinguishing ischemic from hemorrhagic stroke using history and physical examination alone. Methods: This retrospective, observational, case‐control study included 54 adults (27 ischemic, 27 hemorrhagic) with neuroimaging‐confirmed stroke in a tertiary hospital from January to June 2025. De‐identified clinical data were entered into a structured AI prompt. AI predictions were compared with imaging diagnoses, and accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Results: Ischemic stroke patients were slightly older than hemorrhagic cases; hypertension was the most frequent comorbidity. One‐sided weakness was the most common presenting symptom, while headache occurred only in hemorrhagic stroke. Overall accuracy of Google Gemini 2.5 Pro was 66.7% (p = 0.0099). Sensitivity for ischemic stroke was 85.2%, specificity for hemorrhagic stroke was 48.1%, and PPV and NPV were 62.2% and 76.5%, respectively. Conclusion: AI demonstrated moderate diagnostic performance, with greater accuracy for ischemic than hemorrhagic stroke. AI may serve as an adjunct in resource‐limited settings but should not replace neuroimaging. Larger studies and model refinement are recommended.