基于模拟数据的机器学习脑动静脉畸形介入 栓塞术后再通风险预测模型构建及性能比较

Construction and Performance Comparison of Machine Learning-Based Prediction Models for Recanalization Risk Following Interventional Embolization of Brain Arteriovenous Malformations Using Simulated Data

摘要: 本研究旨在构建一种基于模拟数据的机器学习模型，用于预测脑动静脉畸形(AVM)患者术后的再通风险。研究参考了10~20篇近期已发表的相关文献，构建了涵盖患者基本信息、病灶特征、血管构筑特征、临床病史以及治疗信息等多维度模拟数据集。采用Logistic Regression、Random Forest、XGBoost、LightGBM以及Balanced Random Forest这五种机器学习模型进行训练与比较，同时运用SHAP、热图等方法开展特征分析与可视化工作。结果表明，一项基于机器学习算法的研究显示，XGBoost模型在分类问题上的性能评估中表现出色，其AUC值达到0.98，F1分数为0.81，优于其他模型。本研究证实，基于模拟数据构建的机器学习模型能够有效识别AVM栓塞术后再通的高危因素，为临床个性化治疗决策提供可靠的辅助预测工具。

Abstract: This study aims to construct a machine learning model based on simulated data to predict the risk of recanalization after interventional embolization in patients with brain arteriovenous malformations (AVMs). The study references 10 to 20 recently published relevant literature to construct a multi-dimensional simulated dataset covering patient basic information, lesion characteristics, vascular architecture features, clinical history, and treatment information. Five machine learning models—Logistic Regression, Random Forest, XGBoost, LightGBM, and Balanced Random Forest—were used for training and comparison, while SHAP, heatmaps, and other methods were employed for feature analysis and visualization. The results indicate that a study based on machine learning algorithms demonstrates that the XGBoost model excels in the performance evaluation of classification problems, with an AUC value of 0.98 and an F1 score of 0.81, outperforming other models. This study confirms that machine learning models constructed based on simulated data can effectively identify high-risk factors for recanalization after AVM embolization, providing a reliable auxiliary predictive tool for personalized clinical treatment decision-making.