Stroke Outcome Prediction Supported by Deep Learning Algorithm

The STOP-stroke project aims at improving prediction of outcome early after stroke. In order to achieve this, we need to understand reasons (important variables) for prediction in a real clinical prognostication process. We aim to: 1. Test the predictive performance of stroke neurologists for outcome prediction (NIHSS at 24 hours and 3 months and mRS at 3 months after stroke onset) prospectively and in a real clinical setting, and to explore the most important baseline variables in their prognostication process. 2. Test the prediction performance of our DL models when being provided with structured clinical and/or imaging information from the same patients as the neurologists; and to discover most relevant features of the input data. 3. Use the information gained from our experiments for improving our DL algorithm. This will include an error analysis on the missclassifications of models and neurologists to understand the pitfalls of both approaches. We anticipate to develop a robust, reliable and clinically feasible application ready for testing in a prospective, observational trial.