作者：徐守权，唐国文，黄舞标，包仲明，陈锦凤，李超杰，彭美玲，赖瑜梅，文黛薇

单位：

543002广西壮族自治区梧州市中医医院内三科 通信作者：徐守权，E-mail：13907848859@163.com

单位（英文）：

Department of Neurology, Wuzhou Hospital of Traditional Chinese Medicine, Wuzhou 543002, China Corresponding author: XU Shouquan, E-mail: 13907848859@163.com

关键词：

脑梗死; 早期神经功能恶化; BP神经网络; 随机森林; 决策树; 预测;

关键词（英文）：

Cerebral infarction; Early neurological deterioration; BP neural network; Random forest; Decision tree; Forecasting

中图分类号：

R 743.33

DOI：

10.12114/j.issn.1008-5971.2023.00.046

基金项目：

梧州市科学计划项目（202002176）

摘要：

目的 比较BP神经网络、随机森林和决策树预测急性脑梗死（ACI）患者静脉溶栓后发生早期神经功能恶化（END）的效能。方法 选取2021年3月至2022年3月于梧州市中医医院神经内科接受重组组织型纤溶酶原激活剂（rt-PA）静脉溶栓治疗的ACI患者342例，根据静脉溶栓24 h后患者是否发生END将其分为END组（n=66）与非END组（n=276）。比较两组患者临床资料，筛选ACI患者静脉溶栓后发生END的可能影响因素。然后将所有患者按照7∶3的比例分成训练集和测试集，训练集用于构建BP神经网络、随机森林和决策树，测试集用于评估BP神经网络、随机森林和决策树的预测效能。结果 ROC曲线分析结果显示，BP神经网络预测测试集ACI患者静脉溶栓后发生END的AUC为0.957[95%CI(0.918,0.995)]，精确率为0.682，召回率为0.882，灵敏度为0.882，特异度为0.912，正确率为0.912；随机森林预测测试集ACI患者发生END的AUC为0.969[95%CI(0.913,1.000)]，精确率为0.948，召回率为0.989，灵敏度为0.989，特异度为0.925，正确率为0.947；决策树预测测试集ACI患者静脉溶栓后发生END的AUC为0.848[95%CI(0.737,0.959)]，精确率为0.750，召回率为0.883，灵敏度为0.750，特异度为0.914，正确率为0.883。Delong检验结果显示，随机森林预测测试集ACI患者静脉溶栓后发生END的AUC大于决策树（P<0.05）;BP神经网络与决策树、BP神经网络与随机森林预测测试集ACI患者静脉溶栓后发生END的AUC比较，差异无统计学意义（P>0.05）。结论 BP神经网络、决策树及随机森林对ACI患者静脉溶栓后发生END的预测效能良好，其中随机森林对ACI患者静脉溶栓后发生END的区分度优于决策树。

英文摘要：

Objective To compare the efficacy of BP neural network, random forest and decision tree in predicting early neurological deterioration (END) after intravenous thrombolysis in patients with acute cerebral infarction (ACI) . Methods A total of 342 ACI patients who received intravenous thrombolysis with recombinant tissue plasminogen activator (rt-PA) in the Department of Neurology, Wuzhou Hospital of Traditional Chinese Medicine from March 2021 to March 2022 were enrolled. The patients were divided into END group (n=66) and non-END group (n=276) according to the presence or absence of END 24 h after thrombolysis. The clinical data of the two groups were compared to screen the possible influencing factors of END after intravenous thrombolysis in ACI patients. Then all patients were divided into a training set and test set at a ratio of 7∶3. The training set was used to construct the BP neural networks, random forests, and decision trees for predicting the risk of END after intravenous thrombolysis in patients with ACI, and the test set was used to evaluate the predictive efficacy of the BP neural networks, random forests, and decision trees for the risk of END after intravenous thrombolysis in patients with ACI. Results The AUC of BP neural network in predicting END after intravenous thrombolysis in test set ACI patients was 0.957 [95%CI (0.918, 0.995) ] , the precision ratio was 0.682, the recall rate was 0.882, the sensitivity was 0.882, the specificity was 0.912, and the accuracy rate was 0.912. The AUC of random forests in predicting END after intravenous thrombolysis in test set ACI patients was 0.969 [95%CI (0.913, 1.000) ] , the precision ratio was 0.948, the recall rate was 0.989, the sensitivity was 0.989, the specificity was 0.925, and the accuracy rate was 0.947. The AUC of decision tree in predicting END after intravenous thrombolysis in test set ACI patients was 0.848 [95%CI (0.737, 0.959) ] , the precision ratio was 0.750, the recall rate was 0.883, the sensitivity was 0.750, the specificity was 0.914, and the accuracy rate was 0.883. The results of Delong test showed that the AUC of random forests in predicting END after intravenous thrombolysis in test set ACI patients was larger than that of decision tree (P < 0.05) . There was no significant difference in AUC of BP neural network and decision tree, BP neural network and random forest in predicting END after intravenous thrombolysis in test set ACI patients (P > 0.05) . Conclusion BP neural network, decision tree and random forest had good predictive efficacy for END after intravenous thrombolysis in ACI patients, and the discrimination of random forest for END after intravenous thrombolysis in ACI patients is better than that of decision tree

参考文献：