Abstract: Objective To explore the value of head CT perfusion imaging (CTP) parameters combined with triglyceride-glucose (TyG) index and neutrophil-to-lymphocyte ratio (NLR) in evaluating the efficacy of intravenous thrombolysis in patients with acute ischemic stroke (AIS),and to construct a prediction model based on the XGBoost algorithm. Methods A total of 140 AIS patients who received intravenous thrombolysis treatment at Hefei First People's Hospital from February 2023 to October 2025 were retrospectively selected.The patients were divided into the effective group (n=90) and the ineffective group (n=50) based on the change rate of NIHSS score 7 days after thrombolysis.Baseline clinical data and CTP parameters (CBV,CBF,MTT,TTP) of the patients were collected.LASSO regression was used to screen the predictor variables,and the XGBoost algorithm was used to construct a thrombolysis efficacy prediction model,and the importance of model features was explained by the SHAP algorithm. Results The LASSO regression identified six key predictive variables:NLR,TyG,TTP,CBV,CBF,and MTT.The AUC of the XGBoost model in the training set and validation set was 0.900 (95% CI:0.844~0.957) and 0.868 (95% CI:0.784~0.951),respectively,demonstrating excellent discrimination.The calibration curve and clinical decision curve (DCA) indicated that the model had good calibration and clinical practical value.The SHAP analysis showed that the top six feature importance rankings were:MTT,TyG,TTP,NLR,CBV,and CBF. Conclusion The machine learning model constructed based on CT perfusion parameters combined with TyG and NLR indicators and using XGBoost algorithm has shown certain predictive value in the efficacy of intravenous thrombolysis in early AIS patients,aiming to provide a reference for clinical treatment of AIS.

Key words: Acute ischemic stroke, Intravenous thrombolysis, CT perfusion imaging, TyG index, NLR, XGBoost, Machine Learning

摘要： 目的 探讨头颅CT灌注成像(CTP)参数联合甘油三酯-葡萄糖(TyG)指数、中性粒细胞与淋巴细胞比值(NLR)在急性缺血性卒中(AIS)患者静脉溶栓疗效评估中的价值,并基于XGBoost算法构建预测模型。方法 回顾性选取2023年2月至2025年10月合肥市第一人民医院收治的接受静脉溶栓治疗的AIS患者140例。根据溶栓治疗7天后的NIHSS评分变化率将患者分为有效组(n=90)和无效组(n=50)。收集患者基线临床资料及CTP参数(CBV、CBF、MTT、TTP)。采用LASSO回归筛选特征变量,基于XGBoost算法构建溶栓疗效预测模型,并通过SHAP算法对模型特征重要性进行解释。结果 LASSO回归筛选出6个关键预测变量:NLR、TyG、TTP、CBV、CBF、MTT。XGBoost模型在训练集和验证集中的AUC分别为0.900(95%CI:0.844~0.957)和0.868(95%CI:0.784~0.951),显示出优异的区分度。校准曲线和临床决策曲线(DCA)表明模型具有良好的校准度和临床实用价值。SHAP分析显示,特征重要性排名前六位依次为:MTT、TyG、TTP、NLR、CBV、CBF。结论 基于CT灌注参数联合TyG、NLR指标建立的XGBoost构建的机器学习模型在早期AIS患者静脉溶栓疗效上显示出一定的预测价值,旨在为临床治疗AIS提供参考。

关键词: 急性缺血性卒中, 静脉溶栓, CT灌注成像, TyG指数, NLR, XGBoost, 机器学习