基于机器学习算法预测急性缺血性脑卒中静脉溶栓治疗相关的颅内出血转化  

Machine-learning based prediction of hemorrhage transformation associated with thrombolysis of plasminogen activator in acute ischemic stroke patients

作　　者：林亚楠[1] 李焱[2] 韩杰[1] 刘卉馨 LIN Yanan;LI Yan;HAN Jie;LIU Huixin(不详;Department of Neurology,The First Affiliated Hospital of Dalian Medical University,Dalian 116011,China)

机构地区：[1]大连医科大学附属第一医院神经内科,116011 [2]辽宁师范大学化学与生物学交叉研究中心,116029 [3]大连理工大学附属中心医院(大连市中心医院)

出　　处：《中国神经免疫学和神经病学杂志》2025年第2期112-116,124,共6页Chinese Journal of Neuroimmunology and Neurology

摘　　要：目的 构建基于机器学习算法的急性缺血性脑卒中静脉溶栓治疗相关的颅内出血转化(thrombolysis of plasminogen activator related hemorrhage transformation, tPA-HT)预测模型。方法 回顾性分析2010年1月至2022年12月作者医院连续收治的967例接受静脉溶栓治疗的急性缺血性脑卒中患者。根据静脉溶栓后48 h内是否出现颅内出血转化将患者分为tPA-HT组和无tPA-HT组,比较两组患者一般资料、临床资料、实验室检查和影像学检查资料之间的差异,采用随机森林算法构建tPA-HT的机器学习预测模型,采用准确率(accuracy)和受试者工作特征曲线下面积(the area under the receiver-operating characteristic curve, ROC-AUC)评价模型的预测性能,采用部分依赖图(partial dependence plot, PDP)和SHAP(SHapley Additive exPlanations, SHAP)摘要图评价模型的可解释性。结果 (1)tPA-HT组和无tPA-HT组在年龄、房颤病史、白细胞计数、血小板计数、国际标准化比值、美国国立研究院卒中量表(NIHSS)评分、TOAST分型、头CT显示大面积脑梗死征象和大脑中动脉高密度征之间的差异具有统计学意义(均P<0.05);(2)基于29项预测变量构建了tPA-HT机器学习预测模型,其准确率为0.896,ROC-AUC为0.882,模型的预测性能良好;(3)可解释性分析显示,NIHSS评分和头CT大面积脑梗死征象是tPA-HT的主要影响因素。结论 基于机器学习的tPA-HT模型具有良好的预测性能,可以为急性缺血性脑卒中患者静脉溶栓治疗提供更加精准的颅内出血转化风险预测。Objective To predict the risk of thrombolysis of plasminogen activator related hemorrhage transformation(tPA-HT)in patients with acute ischemic stroke using machine-learning based model.Methods We retrospectively enrolled 967 patients who underwent intravenous thrombolysis at our stroke center from January 2010 to December 2022.Based on the occurrence of HT within 48 hours after intravenous thrombolysis,we divided patients into the tPA-HT group and the non tPA-HT group.We compared the differences in clinical data,laboratory tests and imaging findings between the two groups and built the machine learning-based tPA-HT prediction model using random forest algorithm.We tested its prediction performance by the accuracy and the area under the receiver-operating characteristic curve(ROC-AUC).We also analyzed the explainability by the partial dependence plot(PDP)and the SHapley Additive exPlanations(SHAP)summary plot.Results(1)The differences between the tPA-HT and the non tPA-HT groups showed that the age,atrial fibrillation,white blood cell count,platelet count,international normalized ratio,the National Institutes of Health Stroke Scale(NIHSS)score,TOAST subtype,the sign of massive cerebral infarction(MCI),and the hyperdense middle cerebral artery sign(HMCAS)on the CT scan were statistical significances(all P<0.05).(2)The accuracy was 0.896 and the ROC-AUC was 0.882 of the tPA-HT model.(3)NIHSS score and the MCI sign were the two main factors by explainable analysis.Conclusions The machine-learning based tPA-HT model could accurately predict the post-thrombolysis hemorrhage transformation in acute stroke patients.

关 键 词：血栓溶解疗法 出血转化 脑出血 脑缺血 机器学习 可解释性分析 

分 类 号：R743.3[医药卫生—神经病学与精神病学]