复发性外阴阴道假丝酵母菌病患者宫颈病变的多层人工神经网络分类预测模型的构建  

Construction of a multilayer artificial neural network classification and prediction model for cervical lesions in patients with recurrent vulvovaginal candidiasis

作　　者：刘昊[1] 莫坚[1] Liu Hao;Mo Jian(Department of Gynecology,First Peoples Hospital of Nanning,Guangxi Nanning 530022,P.R.China)

机构地区：[1]南宁市第一人民医院妇科,广西南宁530022

出　　处：《中国计划生育和妇产科》2023年第1期77-81,共5页Chinese Journal of Family Planning & Gynecotokology

基　　金：南宁市科学研究与技术开发计划项目(项目编号:20193107)。

摘　　要：目的构建复发性外阴阴道假丝酵母菌病(recurrent vulvovaginal candidiasis,RVVC)患者宫颈病变的多层人工神经网络分类预测模型。方法选取2016年1月至2020年12月在南宁市第一人民医院妇科门诊诊断为RVVC的152例患者,行阴道分泌物检查,并依据HPV E6/E7mRNA检查结果,以HPV E6/E7阳性患者为观察组(65例),HPV E6/E7阴性为对照组(87例)。收集患者相关资料及实验室检查指标,一方面,采用单因素方差分析筛选出与宫颈病变相关的变量作为自变量,纳入多因素Logistic回归模型进行分析;另一方面,随机选取数据集的2/3为训练集(102例),用以建立多层人工神经网络分类预测模型,1/3为测试集(50例),用于该模型测试,采用接受者工作特征(ROC)曲线评估两种模型的预测效能。结果Logistic回归模型预测RVVC发生宫颈病变的ROC曲线下面积为0.833,灵敏度为0.802,特异度为0.808,多层人工神经网络模型预测RVVC发生宫颈病变的ROC曲线下面积为0.987,灵敏度为0.895,特异度为0.889。结论多层人工神经网络分类模型相较于Logistic回归模型,预测效能更高,分类性能优良,结果证实加德纳菌及杂菌、白细胞脂酶是RVVC患者宫颈病变的独立危险因素,乳酸杆菌和过氧化氢是其保护因素。Objective To construct a multilayer artificial neural network classification and prediction model for cervical lesions in patients with recurrent vulvovaginal candidiasis(RVVC).Methods A total of 152 patients diagnosed with RVVC from January 2016 to December 2020 in the Department of Gynecology of the First People’s Hospital of Nanning were selected.Vaginal secretions were examined,and HPV E6/E7 positive patients were selected as the observation group(65 cases)according to the results of HPV E6/E7 mRNA test.The control group(87 cases)was negative for HPV E6/E7.Relevant patient data and laboratory examination indicators were collected.On the one hand,variables related to cervical lesions were screened out by one-way analysis of variance as independent variables and included in multivariate Logistic regression model for analysis.On the other hand,2/3 of the data sets were randomly selected as training sets(102 cases)to establish the classification prediction model of multi-layer artificial neural network,and 1/3 were selected as test sets(50 cases)to test the model.Receiver operating characteristic(ROC)curve was used to evaluate the prediction efficiency of the two models.Results Logistic regression model predicted cervical lesions in RVVC with 0.833 area under ROC curve,0.802 sensitivity and 0.808 specificity;multi-layer artificial neural network model predicted cervical lesions in RVVC with 0.987 area under ROC curve,0.895 sensitivity and 0.889 specificity.Conclusion Compared with Logistic regression model,the multi-layer artificial neural network classification model has higher prediction efficiency and good classification performance.The results confirmed that Gardnerella,hybrid bacteria and leukocyte lipase are independent risk factors of cervical lesions in RVVC patients,and lactobacillus and hydrogen peroxide are protective factors.

关 键 词：复发性外阴阴道假丝酵母菌病 HPV E6/E7mRNA 乳酸杆菌 过氧化氢 

分 类 号：R711.31[医药卫生—妇产科学]