Development and validation of an explainable machine learning prediction model for futile recanalization after mechanical thrombectomy in acute large vessel occlusion stroke

Background Mechanical thrombectomy (MT) is the primary treatment for acute ischemic stroke (AIS) caused by large vessel occlusion (LVO). However, the likelihood of futile recanalization (FR) at 90 days post-MT remains high. Methods This study included 534 AIS patients with anterior circulation LVO who underwent MT, with the primary outcome being FR. The derivation cohort consisted of 445 patients (June 2018–June 2023), while the temporal validation cohort had 89 patients (July 2023–June 2024). The derivation cohort was split into 70% training and 30% internal validation sets. Eleven machine learning (ML) models were trained, tested, and compared, and the best-performing model was selected for optimization and temporal validation. SHapley Additive exPlanations (SHAP) were used for model interpretation. Results The CatBoost model showed the best discriminative ability among the 11 ML models. After feature selection and dimensionality reduction, a final explainable CatBoost model with 12 features was established, accurately predicting FR in both internal (area under the curve (AUC)=0.915) and temporal (AUC=0.930) validations. The model has been deployed as a web application for clinical use. Conclusion We developed a ML prediction model with 12 key features that demonstrates excellent performance in predicting FR. The deployment of this model as a web application offers a promising tool for clinicians to assess FR risk, potentially enhancing patient selection and improving personalized stroke care.