Zhang Hongchen, Lu Chuanhao, Hu Zhen, Sun Deyu, Li Liang, Wu Hongxing, Lu Hua, Lv Bin, Wang Jun, Dai Shuhui, Li Xia
Department of Neurosurgery, Xijing Hospital, The Fourth Military Medical University, Xi'an, China.
Institute for Health Informatics, University of Minnesota, Minneapolis, Minnesota, USA.
CNS Neurosci Ther. 2025 Apr;31(4):e70386. doi: 10.1111/cns.70386.
Although most unruptured intracranial aneurysms (UIAs) have good prognosis after flow diverter (FD) treatment, some remain unoccluded for extended periods, posing a persistent rupture risk. This study aims to develop a predictive model for UIA occlusion after FD treatment through integrating morphological and hemodynamic parameters, which may be critical for personalized postoperative management.
Data from patients with single UIAs treated with stand-alone FD were collected from June 2018 to December 2022 in four cerebrovascular disease centers. Morphological parameters were obtained from 3D reconstructed aneurysm models, and hemodynamic parameters were derived by computational fluid dynamics (CFD) analysis. A predictive model for aneurysm occlusion was constructed using various machine learning algorithms, including logistic regression, Random Forest, XGBoost, and K-Nearest Neighbors. Model performances were evaluated through repeated cross-validation, 0.632 bootstrap, and 0.632+ bootstrap. Shapley additive explanation (SHAP) analysis was employed to assess the contribution of each parameter to UIA occlusion.
Seventy-nine patients were reviewed; a total of 51 cases met the criteria, with an average age of 53.9 ± 9.9 years. The average aneurysm diameter was 3.72 ± 2.72 mm, comprising 29 occlusions and 22 non-occlusions. Five variables were selected for further modeling, including follow-up time > 6 months, aneurysm rupture ratio (ArR), occlusion ratio (OsR), parent artery wall shear stress (WSS), and the change of parent artery WSS. Logistic regression outperformed other algorithms, achieving an area under the curve (AUC) above 0.75, indicating good predictive performance. SHAP analysis revealed that the change of parent artery WSS contributed most significantly to accurate and early prediction. Additionally, a web application software was developed to assist clinicians in real-time aneurysm occlusion prediction.
This study developed a robust predictive model for UIA occlusion following FD treatment by integrating morphological and hemodynamic parameters, which may provide potentially valuable decision-making support for optimizing treatment strategies.
尽管大多数未破裂颅内动脉瘤（UIAs）在血流导向装置（FD）治疗后预后良好，但仍有一些动脉瘤长时间未闭塞，存在持续破裂风险。本研究旨在通过整合形态学和血流动力学参数，建立FD治疗后UIA闭塞的预测模型，这对于个性化术后管理可能至关重要。
2018年6月至2022年12月期间，在四个脑血管疾病中心收集了接受单纯FD治疗的单发UIAs患者的数据。形态学参数从三维重建的动脉瘤模型中获取，血流动力学参数通过计算流体动力学（CFD）分析得出。使用多种机器学习算法构建动脉瘤闭塞预测模型，包括逻辑回归、随机森林、XGBoost和K近邻算法。通过重复交叉验证、0.632自抽样法和0.632+自抽样法评估模型性能。采用Shapley值相加解释（SHAP）分析评估各参数对UIA闭塞的贡献。
共纳入79例患者；51例符合标准，平均年龄53.9±9.9岁。动脉瘤平均直径为3.72±2.72mm，其中29例闭塞，22例未闭塞。选择五个变量进行进一步建模，包括随访时间>6个月、动脉瘤破裂率（ArR）、闭塞率（OsR）、载瘤动脉壁剪切应力（WSS）以及载瘤动脉WSS的变化。逻辑回归算法的表现优于其他算法，曲线下面积（AUC）高于0.75，表明具有良好的预测性能。SHAP分析显示，载瘤动脉WSS的变化对准确和早期预测的贡献最为显著。此外，还开发了一个网络应用软件，以协助临床医生进行实时动脉瘤闭塞预测。
本研究通过整合形态学和血流动力学参数，建立了一个强大的FD治疗后UIA闭塞预测模型，可为优化治疗策略提供潜在有价值的决策支持。
