摘要
目的比较逻辑回归分析、支持向量机和深度神经网络3种机器学习方法在哮喘预测和诊断过程中的准确性。方法选取2020年1月至2021年6月杞县人民医院收治的370例哮喘患者和201例其他呼吸道疾病患者作为研究对象,选择了22条哮喘诊断相关的条件作为输入数据,使用逻辑回归分析、支持向量机和深度神经网络3种算法分别对患者的输入数据进行分析,分析结果与医生专业的临床诊断结果进行比较,计算3种算法在哮喘诊断时的准确率并进行比较。结果当输入全部22个因素时,逻辑回归分析方法诊断哮喘准确率为94%,接收者操作特征曲线下面积(AUC)为0.97;支持向量机方法诊断哮喘准确率为82%,AUC为0.83;深度神经网络方法诊断哮喘准确率为98%,AUC为0.99。结论深度神经网络能够根据患者的临床数据准确预测哮喘,该深度神经网络的方法预测哮喘时比逻辑回归分析和支持向量机方法准确率更高,可以用于医生诊断哮喘时的助理服务,这一研究可以提高医生诊断哮喘时的准确性,值得在临床推广应用。
Objective To compare the accuracy of logistic regression analysis,support vector machine and deep neural network in the prediction and diagnosis of asthma.Methods From January 2020 to June 2021,370 cases of 201 cases patients with asthma and other respiratory disease patients in Qi County People’s Hospital as the research object,22 asthma diagnosis related conditions were selected as input data.Logistic regression analysis,support vector machine and deep neural network algorithms were used to analyze the input data of patients,and the analysis results were compared with the clinical diagnosis results of doctors.The accuracy of the three algorithms in the diagnosis of asthma were calculated and compared.Results When all 22 factors were input,the accuracy of logistic regression analysis was 94%and the area under the curve(AUC)was 0.97.The accuracy of support vector machine was 82%and AUC was 0.83.The accuracy of deep neural network was 98%and AUC was 0.99.Conclusion The depth of the neural network can predict the asthma according to patients’clinical data.The accuracy of depth of the neural network method to predict asthma were higher than logistic regression analysis and support vector machine method,and it can be used for doctors to diagnose asthma of the assistant service.This research can improve the accuracy of the doctors to diagnose asthma,and it is worth popularizing in clinical application.
作者
刘宁
LIU Ning(Department of Respiratory Medicine,Qi County People’s Hospital,Kaifeng 475200,China)
出处
《河南医学研究》
CAS
2021年第35期6560-6563,共4页
Henan Medical Research
关键词
哮喘
逻辑回归分析
支持向量机
深度神经网络
asthma
logistic regression analysis
support vector machine
deep neural network