Multimodal Stroke Outcome Prediction

This project aims to develop and validate a multimodal artificial intelligence model that can predict recovery and psychological outcomes in patients following acute ischemic stroke. Patients admitted within 24 hours of stroke onset will be prospectively followed for 12 months. The model will combine information from four complementary sources: baseline clinical characteristics, whole-brain non-contrast CT images, quantitative EEG measurements, and objective movement data collected using an in-house monitoring device. These data will be used to predict functional outcomes such as the modified Rankin Scale, NIHSS, and Barthel Index, as well as post-stroke depressive symptoms measured using the Beck Depression Inventory at 3, 6, and 12 months. The main objective is to determine whether combining these different types of information can provide more accurate and comprehensive predictions of individual recovery than conventional clinical information alone. The study will also examine the contribution of each modality and develop an interpretable model that could eventually support personalized rehabilitation planning and post-stroke care.