Machine Learning for Predicting Thromboembolic Events Following Flow Diverter Treatment of Intracranial Aneurysms: A Multicenter Retrospective Study

Neurol Ther (2025) 14:2171–2185 https://doi.org/10.1007/s40120-025-00808-9 ORIGINAL RESEARCH Machine Learning for Predicting Thromboembolic Events Following Flow Diverter Treatment of Intracranial Aneurysms: A Multicenter Retrospective Study Yunpeng Lin · Xiaoning Liu · Bingcheng Ren · Jiwen Wang · Yang Li · Xiangbo Liu · Yidi Wang · Fushun Xiao · Shiqing Mu Received: June 18, 2025 / Accepted: July 23, 2025 / Published online: August 27, 2025 © The Author(s) 2025 ABSTRACT Introduction: Flow diverters (FD) have gradually become the preferred treatment option for complex and large intracranial aneurysms. Postoperative thromboembolic events (TEEs) are among the most common complications associated with endovascular treatment. However, widely applicable predictive tools for the occurrence of TEEs are currently lacking. Methods: This retrospective study included clinical data from 377 patients (a total of 451 Yunpeng Lin, Xiaoning Liu and Bingcheng Ren have contributed equally to this work. Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s40120-025-00808-9. aneurysms) treated with flow diverters at two neurointerventional centers between June 2018 and September 2022. Thirty-nine baseline patient characteristics were included as clinical variables. The primary endpoint was the occurrence of postoperative ischemic events. The dataset was randomly divided into a training set (80%) and a testing set (20%). We performed fivefold cross-validation and applied Lasso regression to the training set to identify the most informative features. Multiple machine learning (ML) algorithms were employed to construct predictive models. Model performance was evaluated on the testing set using the area under the receiver operating characteristic curve (AUC-ROC), the area under the precision–recall curve (AUC-PR), and calibration plots. SHapley Additive exPlanations (SHAP) analysis was used Y. Lin · X. Liu · B. Ren · J. Wang · Y. Li · X. Liu · Y. Wang · F. Xiao (*) Department of Neurosurgery, Tianjin Medical University General Hospital, Tianjin 300052, China e-mail: Y. Li e-mail: Y. Lin e-mail: Y. Wang e-mail: X. Liu e-mail: S. Mu (*) Department of Interventional Neuroradiology, Beijing Neurosurgical Institute, Beijing Tiantan Hospital, Capital Medical University, Beijing 100070, China e-mail: B. Ren e-mail: J. Wang e-mail: X. Liu e-mail: Vol.:(0123456789) 2172 to visualize feature contributions and to interpret individual case predictions. Results: Among 377 patients, 21 (5.6%) experienced TEEs. A machine learning model incorporating 10 variables was developed, with the support vector machine (SVM) model demonstrating the best performance—achieving an AUC-ROC of 0.96 and an AUC-PR of 0.88 in validation. The key predictive factors included aneurysm width, low-density lipoprotein (LDL) levels, hypertension, aneurysm location, triglycerides (TG), and diabetes. Additionally, a webbased tool was developed to assist clinicians in applying the model in practice. Conclusions: We developed a machine learning model to predict the risk of TEEs following FD implantation for intracranial aneurysms, and demonstrated its clinical potential through internal validation. This tool can assist neurointerventionalists in estimating the probability of TEE occurrence based on patient clinical data and aneurysm characteristics, enabling the development of personalized treatment strategies. Keywords: Intracranial aneurysm; Flow diversion; Thromboembolic events; Machine learning; Predictive models Key Summary Points Why carry out this study? Flow diverters (FD) have become a mainstay for complex intracranial aneurysms; however, thromboembolic events (TEEs) remain a major postoperative challenge despite routine dual antiplatelet therapy (DAPT). The currently available tools for anticipating TEE risk remain inadequate, largely because conventional statistical models fail to account for complex, nonlinear interdependencies among risk variables. This study aims to construct an automated machine learningbased model for predicting the likelihood of TEEs after treatment. Neurol Ther (2025) 14:2171–2185 What was learned from the study? This study developed a high-performance machine learning model based on support vector machine to predict the risk of TEEs following FD implantation, demonstrating robust predictive capability. Using SHapley Additive exPlanations analysis, key predictive factors such as aneurysm width, LDL levels, hypertension, and aneurysm location were identified. These findings may facilitate optimized allocation of medical resources and advance the implementation of precision medicine in the management of cerebrovascular diseases. INTRODUCTION Intracranial aneurysms are common vascular abnormalities, with a prevalence of approximately 1–8% in the adult population. Aneurysmal rupture is the leading cause of non-traumatic subarachnoid hemorrhage. Because of its high mortality and morbidity rates, it imposes a significant economic burden [1]. The treatment of complex, large, and wide-necked aneurysms remains particularly challenging. The advent of flow diverters has introduced a safer and more effective therapeutic option, and they are increasingly regarded as the preferred approach for managing such aneurysms [2]. Thromboembolic events are the most common complications following endovascular treatment of intracranial aneurysms. Due to their relatively high porosity and up to 55% aneurysm wall coverage, FDs are associated with a TEE incidence ranging from 3.6% to 10% [3–6]. A dual antiplatelet therapy (DAPT) regimen is routinely administered following FD placement to prevent such adverse events. Despite the perioperative use of DAPT, standardized anticoagulation protocols, and platelet function monitoring, the occurrence of TEEs during both the acute phase and follow-up after FD implantation remains a major clinical challenge. ML leverages computational efficiency and large-scale data to rapidly analyze Neurol Ther (2025) 14:2171–2185 complex information, offering researchers novel approaches for developing experimental methodologies and identifying promising research questions [7]. In recent years, ML has been increasingly applied in the medical field to aid in disease diagnosis and surgical outcome prediction. Unlike traditional multivariate regression methods, which are often limited to linear or nonlinear relationships between variables, ML offers a more flexible and nuanced analysis of complex interactions [8]. In this study, we evaluated six ML algorithms—logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), multilayer perceptron (MLP), and k-nearest neighbors (KNN)—and compared their performance in predicting postoperative ischemic events following FD treatment for intracranial aneurysms. The goal was to develop a reliable prediction tool that inte (...truncated)