Projects & Grants
| Development and Validation of Adaptive Explainable AI Machine Learning Model for predicting a Favorable Clinical Outcome (mRS 0-2) Before and After Mechanical Thrombectomy in Ischemich Stroke Patents. | |
|---|---|
| Project Id | SGS18/LF/2026 |
| Main solver | prof. MUDr. Václav Procházka, Ph.D., MSc., MBA |
| Period | 1/2026 - 12/2026 |
| Provider | Specifický VŠ výzkum |
| State | solved |
| Anotation | This project develops an adaptive machine learning prognostic model for predicting a chance of a favourable 3-month functional outcome (mRS 0-2) from pre- and post-procedural data of a monocentric retrospective patient cohort with large-vessel occlusion stroke after mechanical thrombectomy. The project establishes a fully reproducible data-processing and modeling framework that incorporates clinically validated cleaning rules, structured handling of missingness, and monotonicity-constrained gradient boosting (LightGBM). The model integrates demographic, clinical, imaging, and treatment-timing features that are available at the time of intervention decision-making, enabling individualized real-time prognostic estimation. The project follows TRIPOD principles by applying a stratified temporal split into development and validation sets and assessing model quality through discrimination, calibration accuracy, and clinical utility. Evaluation includes ROC AUC, Brier score, logistic calibration with slope and intercept, and decision-curve analysis to quantify net clinical benefit. Model interpretability is provided through SHAP-based explanations, including subgroup-specific analyses such as age-stratified feature importance. The project also develops a secure, interactive Streamlit web interface that allows clinicians and researchers to explore model behavior and generate patient-specific prognostic estimates using clinically consistent input logic and faithful categorical encoding. This work establishes a transparent, extensible, and highly interpretable prognostic framework for real-time decision support in endovascular stroke therapy and provides the necessary foundation for future multicenter prospective validation. |
| Total Costs | 165 000 CZK |
