中国医学前沿杂志（电子版）2025，Vol.17Issue(11)：49-58,10.DOI:10.12037/YXQY.2025.11-07
脑梗死溶栓危险因素及预后的人工智能模型建立
Establishment of an artificial intelligence model for predicting risk factors and prognosis in cerebral infarction after thrombolysis
摘要
Abstract
Objective This study aims to identify the risk factors for intravenous thrombolysis in patients with cerebral infarction and to establish a clinical prognostic prediction model.Methods Patients with cerebral infarction admitted to our hospital from January to December 2023 were included.Their clinical characteristics,laboratory examination results,and imaging data were recorded.The entire dataset was randomly divided into a training set and a testing set at a 7︰3 ratio.The training set was used for variable selection and model development,while the testing set was used for model validation.Using Python,we established an XGBoost model,a LightGBM model,a decision tree model,an artificial neural network(ANN)model,a support vector machine(SVM)model,and a Bayesian network(BN)model.Results A total of 227 patients were included in the study,among whom 51 had a poor prognosis.The training set contained 158 patients(35 unfavorable)and the test set 69 patients(16 unfavorable).This study employs 5-fold cross-validation to optimize model parameters on the training set and evaluates the final performance on an independent test set.The cross-validation results of the training set show that the XGBoost model algorithm performs the best,with an average area under the curve(AUC)of 0.87±0.07,which outperforms other models(LightGBM:0.86±0.08,ANN:0.86±0.05,SVM:0.85±0.07,decision tree:0.84±0.07,BN:0.82±0.03).On the independent test set,XGBoost model also maintained the best performance,with an AUC of 0.79(95%CI:0.65-0.93),an accuracy of 0.71,an precision of 0.43,a recall of 0.75,and an F1 score of 0.55.The AUCs of other model test sets are as follows:LightGBM 0.80,ANN 0.82,decision tree 0.69,BN 0.70,and SVM 0.79.Combining the results of the training set and the test set,XGBoost model demonstrates the best and most robust prediction performance.To address the modeling challenges with a limited sample size,this study prioritizes the feature importance assessment of the XGBoost model for initial feature screening and explores the final predictors based on this.The feature importance analysis based on XGBoost model shows that the top 5 most important features for prognosis prediction are,in order:National Institutes of Health stroke scale(NIHSS)score at admission,TOAST classification,C reactive protein(CRP),creatine kinase myocardial band(CK-MB)and creatine kinase(CK).Conclusions Through systematic comparison,the XGBoost model algorithm yielded the best-performing and most robust prediction model on this dataset.Patients with acute cerebral infarction undergoing intravenous thrombolysis,NIHSS score,TOAST classification,CRP,CK-MB and CK at admission are potential key prognostic predictors.The research results provide new ideas and algorithm candidates for the risk stratification of stroke prognosis.关键词
脑梗死/人工智能/临床预测模型/静脉溶栓/机器学习Key words
Cerebral infarction/Artificial intelligence/Clinical prediction model/Intravenous thrombolysis/Machine learning引用本文复制引用
孟琦,于佩佩,崔艺浓,李娜,庞艳梅..脑梗死溶栓危险因素及预后的人工智能模型建立[J].中国医学前沿杂志（电子版）,2025,17(11):49-58,10.基金项目
保定市科技计划资助项目(2241ZF248) S&T Program of Baoding(2241ZF248) （2241ZF248）
