基于人工智能的儿童甲流和乙流辅助诊断模型研究

Research on a child influenza A and B auxiliary diagnosis model based on artificial intelligence

目的通过机器学习算法进行甲流和乙流阳性感染患儿的智能辅助诊断模型研究,协助开展传染病门诊预诊。方法以2013年1月—2020年6月在复旦大学附属儿科医院门诊就诊的呼吸道感染性疾病患儿为研究对象,纳入基本信息、鼻咽拭子及血常规检验数据,采用Python进行数据处理和统计分析,基于Logistics回归和GBDT模型构建辅助诊断模型并计算特征值,以ROC曲线、AUC值和模型概率预测箱型图等指标作为模型性能判断标准。结果经鼻咽拭子确诊为单甲流阳性38094例,单乙流阳性24792例,甲乙流合并215例,共计63101例。共纳入25个指标作为模型特征值。基于Logistics模型和GBDT模型构建的甲流辅助诊断模型AUC值分别为0.877和0.884,前5位重要特征为年龄、单核细胞百分比、白细胞、淋巴细胞绝对值和C反应蛋白;乙流模型AUC值分别为0.895和0.902,前5位特征为年龄、单核细胞百分比、嗜酸性细胞计数、白细胞和血小板。GBDT效果均好于Logistics模型,且在鉴别单乙流阳性病例时性能最佳(AUC=0.902)。结论本研究建立起基于血常规检验数据的儿童甲乙型流感AI辅助诊断模型,可在诊前较为准确地从呼吸道感染性疾病人群中识别甲流和乙流感染阳性患儿,迁移性好,能够在实际应用中发挥诊前辅助诊断作用。

Objective To conduct an intelligent auxiliary diagnosis model of influenza in children(Flu A and Flu B)was conducted thought machine learning algorithm,so as to assist in the pre-diagnosis of infectious diseases.Methods Taking the children with respiratory tract infectious who were in the outpatient clinic from Jan 2013 to Jun 2020 as the research object,the basic characteristics information,nasopharyngeal swabs,and routine blood test data were included,and Python was used for data processing and statistical analysis.Then,based on the machine learning algorithm Logistics regression model and the GBDT model to construct an auxiliary diagnostic model and calculate the eigenvalues.The indicators such as ROC,AUC value and model probability prediction box plot were used as the criteria to judge the performance of models.Results Among the scope of the study,nasopharyngeal swabs showed that 38094 cases were positive for Flu A infection,24792 cases were positive for Flu B infection,and 215 cases were positive for combined Flu A with Flu B infection,totaling 63101 cases.Twenty-five indicators were included as the model characteristic values.The AUC values of Flu A auxiliary diagnosis model based on Logistics model and GBDT model were 0.877 and 0.884,respectively,and the first five crucial characteristics were age,percentage of monocytes,white blood cells,lymphocytes absolute value and Creactive protein.The AUC values of Flu B auxiliary diagnosis were 0.895 and 0.902,and the top five important characteristics were age,percentage of monocytes,eosinophilic cell count,white blood cells and platelets.The effects of GBDT model are better than that of Logistics model,and GBDT model has the best performance in the differential diagnosis of positive cases of single Flu B infection(AUC=0.902).Conclusion In this study,an intelligent auxiliary diagnosis model of Flu A and Flu B in children based on blood routine test was established,which could accurately identify positive patient with Flu A and Flu B from the patient with respiratory tract infectious diseases before diagnosis.With good migration,it could play a role on the auxiliary diagnosis before diagnosis in practical application scenarios.