Wake-up Stroke Outcome Prediction by Interpretable Decision Tree Model

Outcome prediction in wake-up ischemic stroke (WUS) is important for guiding treatment strategies, in order to improve recovery and minimize disability. We aimed at producing an interpretable model to predict a good outcome (NIHSS 7-day<5) in thrombolysis treated WUS patients by using Classification and Regression Tree (CART) method. The study encompassed 104 WUS patients and we used a dataset consisting of demographic, clinical and neuroimaging features. The model was produced by CART with Gini split criterion and evaluated by using 5-fold cross-validation. The produced decision tree model was based on NIHSS at admission, ischemic core volume and age features. The predictive accuracy of model was 86.5% and the AUC-ROC was 0.88. In conclusion, in this preliminary study we identified interpretable model based on clinical and neuroimaging features to predict clinical outcome in thrombolysis treated wake-up stroke patients.