Predicting 12 Month Functional Recovery After Ischemic Stroke Using Lesional Topography-Based Deep Neural Network

Motivation: Precise prediction of functional recovery after ischemic stroke remains challenging. Current models, primarily based on clinical characteristics and lesion volume, often lack the spatial lesion information essential for accurate prognosis. Goal(s): This study aims to improve the prediction of favorable outcomes at one-year post-stroke by developing a model that integrates topographic features into a deep neural network (DNN). Approach: A DNN model was trained on clinical, volume, and topographic features to assess prediction accuracy, sensitivity, and specificity. Results: Adding topographic features enhanced model accuracy to 93.8% compared to models using only clinical or clinical and volume features. Impact: This study demonstrates that incorporating topographic features into predictive models provides additional information, significantly improving outcome predictions for ischemic stroke patients.