Prediction of Recurrence of Intracranial Aneurysms Treated with Woven EndoBridge Device

Introduction: Endovascular treatment of intracranial aneurysms using the Woven EndoBridge (WEB) device has shown promise, but predictors of recurrence remain poorly studied. This study aimed to develop interpretable machine learning models to identify risk factors associated with long-term aneurysm recurrence after WEB deployment.

Methods: A large retrospective multicenter study was conducted using data on demographics, aneurysm characteristics, treatment details, and clinical outcomes. The main outcome was defined as aneurysm recurrence represented by worsening aneurysm filling from initial Raymond-Roy grade 1 to final follow-up Raymond-Roy grade 2-3 or initial Raymond-Roy grade 2 to final Raymond-Roy grade 3. Multiple machine learning classification algorithms including Naive Bayes, Logistic Regression, K-Nearest Neighbors, Random Forest, AdaBoost, and CatBoost were trained on 75% of the data and validated on the remaining 25% using stringent 10-fold cross-validation. Hyperparameter tuning was performed to optimize model performance. Evaluation metrics included area under the ROC curve (AUC), accuracy, F1 score, precision and recall. The best performing model was selected and interpreted using SHAP values to explain the influence of individual features on predictions.

Results: Out of 1218 cases included, aneurysm recurrence occurred in 10% of cases. The optimal CatBoost model achieved an AUC of 0.65, accuracy of 90%, and F1 score of 0.71 on the test set. The most influential features for predicting recurrence were ruptured presentation, WEB single-layer (SL), aneurysm location of Acom, height, presence of multiple aneurysms, and partial aneurysm thrombosis.

Conclusion : This machine learning modeling study identified several key aneurysm morphological and procedural risk factors associated with increased recurrence after WEB treatment. The developed predictive models can assist physicians in individualized risk assessment and inform clinical decision-making regarding optimal patient selection to maximize durable occlusion rates, such as different techniques or approaches.