| 摘要: |
| ［摘要］ 目的 基于动脉瘤形态学、血流动力学和一般临床指标，通过逻辑回归（LR）和极端梯度提升（XGBoost）方法构建开颅夹闭术中颅内动脉瘤（IA）破裂的预测模型，并对其预测效能进行验证和对比。方法 招募2020年1月至2025年4月邢台市中心医院收治的IA患者330例，以7∶3比例随机将其分为训练集（231例）与验证集（99例）。所有患者均接受开颅夹闭术治疗，根据术中IA发生破裂情况将其分为破裂组和未破裂组。于术前通过计算机断层扫描血管造影（CTA）检查，获取动脉瘤形态学指标以及血流动力学指标，并收集患者一般临床资料。分别采用LR和XGBoost方法构建开颅夹闭术中IA破裂的预测模型，采用受试者工作特征（ROC）曲线和校正曲线对两种模型的预测效能进行评估。结果 在训练集与验证集中，破裂组高血压、颅内动脉粥样硬化占比、系统炎症反应指数（SIRI）、瘤体最大径、纵横比（AR，瘤体长径/瘤颈宽度）、尺寸比（SR，瘤体最大径/载瘤动脉直径）、壁面剪切力梯度（WSSGA）及震荡剪切指数（OSI）均高于未破裂组，标准化表面最大剪切力（NWSSM）低于未破裂组，差异有统计学意义（P<0.05）。基于训练集数据，LR分析结果显示，高血压、颅内动脉粥样硬化、SIRI、AR、SR、OSI、NWSSM是开颅夹闭术中IA破裂的影响因素（P<0.05）。预测模型方程：Logit（P）=-1.215+1.096×高血压+1.329×颅内动脉粥样硬化+0.629×SIRI+0.707×AR+0.719×SR+0.838×OSI-0.833×NWSSM。XGBoost方法构建的预测模型中，特征重要性排序前6位因素依次为OSI、SIRI、SR、NWSSM、AR和颅内动脉粥样硬化。基于训练集数据的ROC曲线分析结果显示，XGBoost模型和LR模型均可有效预测开颅夹闭术中IA破裂的发生情况［AUC（95%CI）=0.915（0.875～0.955）；AUC（95%CI）=0.838（0.780～0.897）］。校正曲线分析结果显示，XGBoost模型和LR模型的校准度分别为0.885和0.828，一致性指数分别为0.879和0.812。基于验证集数据的效能分析结果也显示，XGBoost模型较LR模型具有更好的预测效能和准确性。结论 基于OSI、SIRI、SR、NWSSM、AR和颅内动脉粥样硬化等6个指标的XGBoost模型可有效预测开颅夹闭术中IA破裂的发生情况，有助于临床医师及早识别高风险患者，改善患者预后。 |
| 关键词: 颅内动脉瘤 开颅夹闭术 动脉瘤破裂 预测模型 逻辑回归 极端梯度提升 |
| DOI：10.3969/j.issn.1674-3806.2026.03.13 |
| 分类号:R 651.1 |
| 基金项目:邢台市重点研发计划自筹项目（编号：2024ZC194） |
|
| Construction of a model for predicting intracranial aneurysm rupture during craniotomy and clipping based on aneurysm morphological, hemodynamic and general clinical indicators and verification of its predictive efficacy |
|
WANG Shengjun, FENG Yumin, LI Yan, HAO Jinmin
|
|
Department of Neurosurgery, Xingtai Central Hospital, Xingtai 054000, China
|
| Abstract: |
| ［Abstract］ Objective To construct models for predicting intracranial aneurysm(IA) rupture during craniotomy and clipping based on aneurysm morphological, hemodynamic and general clinical indicators using logistic regression(LR) and extreme gradient boosting(XGBoost) methods, and to verify and compare the predictive efficacy of the models. Methods A total of 330 patients with IA who were admitted to Xingtai Central Hospital from January 2020 to April 2025 were recruited and randomly divided into training set(231 patients) and verification set(99 patients) at a ratio of 7∶3. All the patients received craniotomy and clipping, and were divided into rupture group and non-rupture group according to whether or not IA ruptured during the operation. Before the operation, computed tomography angiography(CTA) was performed on the patients to obtain the aneurysm morphological indicator and hemodynamic indicator, and the general clinical data of the patients were collected. The models for predicting IA rupture during craniotomy and clipping were constructed by using the LR and XGBoost methods, respectively. The predictive efficacy of the two models was evaluated by using the receiver operating characteristic(ROC) curve and calibration curve. Results In the training set and the verification set, the proportions of hypertension and intracranial atherosclerosis, and the systemic inflammatory response index(SIRI), the maximum diameter of the aneurysm, the aspect ratio(AR, the long diameter of the aneurysm/the width of the aneurysm neck), and the size ratio(SR, the maximum diameter of the aneurysm/the diameter of the parent artery), the wall shear stress gradient average(WSSGA), and the oscillatory shear index(OSI) in the rupture group were greater than those in the non-rupture group, while the normalized wall shear stress maximum(NWSSM) in the rupture group was less than that in the non-rupture group, and the differences were statistically significant(P<0.05). Based on the data in the training set, the results of LR analysis showed that hypertension, intracranial atherosclerosis, SIRI, AR, SR, OSI and NWSSM were the influencing factors of IA rupture during craniotomy and clipping(P<0.05). The equation of the prediction model: Logit(P)=-1.215+1.096×hypertension+1.329×intracranial atherosclerosis+0.629×SIRI+0.707×AR+0.719×SR+0.838×OSI-0.833×NWSSM. In the prediction model constructed by the XGBoost method, the top 6 factors in terms of the importance of features were OSI, SIRI, SR, NWSSM, AR and intracranial atherosclerosis in sequence. The results of ROC curve analysis based on the training set data showed that both the XGBoost model and the LR model could effectively predict the occurrence of IA rupture during craniotomy and clipping［AUC(95%CI)=0.915(0.875-0.955); AUC(95%CI)=0.838(0.780-0.897)］. The results of calibration curve analysis showed that the calibration degrees of the XGBoost model and the LR model were 0.885 and 0.828, respectively, and the consistency indices were 0.879 and 0.812, respectively. The analytical results of the predictive efficacy of the models based on the verification set data also showed that the XGBoost model had better predictive efficacy and accuracy than the LR model. Conclusion The XGBoost model based on 6 indicators including OSI, SIRI, SR, NWSSM, AR and intracranial atherosclerosis can effectively predict the occurrence of IA rupture during craniotomy and clipping, which helps clinicians identify high-risk patients early and improve their prognosis. |
| Key words: Intracranial aneurysm(IA) Craniotomy and clipping Aneurysm rupture Prediction model Logistic regression(LR) Extreme gradient boosting(XGBoost) |
