Abstract Number: Esoc2026a1658 a Machine Learning-Based Predictive Model for Assessing the Risk of Cerebral Microbleeds in Ischemic Stroke Patients

Abstract Background and aims Cerebral microbleeds (CMBs) are common in ischemic stroke and have been linked to hemorrhagic transformation, recurrence, and cognitive decline, yet they often remain undetected. Using routine clinical variables, we developed a machine-learning model that estimates risk of CMBs in ischemic-stroke patients, aiming to provide an early signal that could inform discussions on antithrombotic choice and, in turn, contribute to better patient outcomes. Methods In this prospective multicenter study, consecutive ischemic-stroke patients admitted to 19 hospitals in Shaanxi, China, were enrolled from February 2022 to October 2024. The dataset was randomly split into training (70%) and validation (30%) cohorts. Predictive features were selected using LASSO regression, and twelve algorithms (nine machine-learning and three deep-learning models) were trained. The best-performing model was identified by comparing AUC, accuracy, precision, recall and F1-score in the validation cohort, and SHapley Additive exPlanation (SHAP) was applied to interpret the features of the optimal machine-learning model. Results A total of 1,284 ischemic-stroke patients were prospectively enrolled. Among twelve trained models, the Naïve Bayes classifier achieved the highest discrimination in the validation cohort (AUC 0.71, 95 % CI 0.66–0.76; accuracy 0.65; recall 0.79; F1-score 0.68) and was selected as the final tool. An open-access web application was deployed to provide real-time CMB risk estimates at the bedside. Conclusions This study developed and validated a machine-learning model using routinely available clinical variables to predict the risk of cerebral microbleeds in ischemic-stroke patients. The model enables early identification of high-risk patients at baseline, supporting individualized antithrombotic treatment decisions. Conflict of interest Xiao Zhang; Luojun Wang; Dong Wei; Wen Jiang: nothing to disclose