中国全科医学2018，Vol.21Issue(12)：1413-1418,6.DOI:10.3969/j.issn.1007-9572.2017.00.189
基于BP神经网络的急性脑梗死患者自发性出血性转化的风险预测研究
Prediction of Risk of Spontaneous Hemorrhagic Transformation in Patients with Acute Cerebral Infarction Based on Back Propagation Neural Networks
摘要
Abstract
Objective To predict the risk of spontaneous hemorrhagic transformation(HT) in patients with acute cerebral infarction(ACI) with the back propagation(BP) neural network,to provide evidence for the prevention and treatment of spontaneous HT.Methods A total of 372 ACI patients who were hospitalized but did not receive thrombolytic therapy in Affiliated Hospital of North China University of Science and Technology and Tangshan Gongren Hospital from January 2014 to January 2017 were enrolled.All subjects were assigned into the case group(n=124) or the control group(n=248) according to the development of HT as revealed by imaging examinations.The medical records were retrospectively collected,including the basic information regarding previous history,clinical data,laboratory examinations,and imaging data.All parameters with statistical significance in univariate analyses were included in the Logistic regression model and the BP neural network model based risk prediction model,and the receiver operating characteristic(ROC) curve were calculated to compare the performance of the two models for risk prediction.Results The proportion of patients with a medical history of atrial fibrillation,the National Institute of Health Stroke Scale(NIHSS) score,the rate of patients with a history of anticoagulant therapy,prothrombin time(PT), white blood cell(WBC) count,fibrinogen level,rate of patients with large area cerebral infarction,rate of patients with early low density shadows on computed tomography(CT) scans,and the incidence of leukoaraiosis(LA) were higher in the case group than in the control group(P<0.05).A lower rate of patients with a history of anticoagulant therapy,rate of patients with a history of antiplatelet therapy(APT),high density lipoprotein cholesterol(HDL-C) concentration,and albumin level(ALB) were found in the case group than in the control group(P<0.05).The expression form of the Logistic regression model was Logit (P)=0.109×NIHSS score+1.380×PT+0.355×WBC+1.320×LA-0.842×APT-1.144×HDL-C-0.087×ALB-17.554.The number of the hidden layers was 1 in the BP neural network model,and the number of neurons was 5 in the hidden layer.The area under the ROC curve of the BP neural network model(0.969) was greater than that(0.906) of the Logistic regression model for predicting the risk of spontaneous HT in ACI patients(Z=3.601,P<0.001).Sensitivity analyses of BP neural network model showed that the factors that significantly affected the development of spontaneous HT included PT(100.0%),ALB level(75.8%), WBC count(75.8%),NIHSS score(52.4%),the presence of early low density shadows on CT scans(36.8%),HDL-C concentration(35.4%),a history of APT(33.6%),development of massive cerebral infarction(31.9%),fibrinogen level(23.1%), a medical history of atrial fibrillation(22.1%),a history of using anticoagulant drugs(20.3%),a history of anticoagulant therapy(18.1%),and development of LA(13.6%).Conclusion The BP neural network model is highly effective for the prediction of the risk of spontaneous HT in ACI patients,which may be used to formulate the assisted clinical decision making.关键词
脑梗死/自发性出血性转化/神经网络(计算机)/预测Key words
Brain infarction/Spontaneous hemorrhagic transformation/Neural network(computer)/Forecasting分类
医药卫生引用本文复制引用
汪可可,王国立,武建辉,周莹,彭延波,曹英志,宋宇,张晓雅,袁欣,王倩..基于BP神经网络的急性脑梗死患者自发性出血性转化的风险预测研究[J].中国全科医学,2018,21(12):1413-1418,6.基金项目
河北省高等学校科学技术研究项目(QN2017349) （QN2017349）
