BP神经网络在以血尿为主要表现患儿肾脏病理预测中的应用

目的探讨不同病理类型以血尿为主要表现的患儿各临床指标的差异,并建立基于临床资料的BP神经网络预测模型。方法收集2003年6月至2018年12月重庆医科大学附属儿童医院以血尿为主要表现并行肾活检患儿的临床资料及肾脏病理结果,进行各临床指标的差异性比较,并建立以血尿为主要表现患儿肾脏病理的BP神经网络预测模型。结果共纳入438例患儿。其中男232例,女206例;起病年龄(7.00±3.15)岁。按照不同临床表现分为:镜下血尿组(179例)、肉眼血尿组(81例)、镜下血尿并蛋白尿组(44例)、肉眼血尿并蛋白尿组(134例)。差异性检验结果显示,各组性别、起病年龄、病程、诱因、尿爱迪(Addis)计数中红细胞数、24 h尿蛋白定量、血清素氮、血肌酐、血清蛋白、血IgA水平差异均有统计学意义(均P<0.05);不同病理类型性别、起病年龄、病程、家族史、尿Addis计数中红细胞数、24 h尿蛋白定量、血尿素氮、血肌酐、血清蛋白、血IgA、C3水平差异均有统计学意义(均P<0.05)。基于以上指标,构建BP神经网络预测模型,并通过留一法验证了该预测模型的准确率为61.19%。结论通过进行不同临床表现和病理分型下各指标的差异性比较,建立以血尿为主要表现患儿的肾脏病理的BP神经网络预测模型。通过相关指标能较为准确预测肾脏病理,为肾活检时机提供依据。

Application of BP neural network in renal pathological prediction in children with hematuria as main clinical manifestation

Objective To explore the differences in clinical indicators of different pathological types of children with hematuria as the main manifestation,and to establish a BP neural network prediction model based on clinical data.Methods The clinical data and renal pathological results of children who were referred to Children′s Hospital of Chongqing Medical University from June 2003 to December 2018 for evaluation of hematuria as the main manifestation were collected,the significant differences in these clinical indicators were analyzed,and a BP neural network model for predicting renal pathology in children with hematuria as the main manifestation was established.Results A total of 438 cases were enrolled in this study,including 232 males and 206 females,with the onset age of(7.00±3.15)years old.According to different clinical manifestations,the children were divided into microscopic hematuria group(179 cases),gross hematuria group(81 cases),microscopic hematuria and proteinuria group(44 cases),and gross hematuria and proteinuria group(134 cases).There were significant differences in sex ratio,onset age,course of disease,inducement,Addis count of urinary red cells,24-hour proteinuria,blood urea nitrogen,serum creatinine,serum albumin and serum IgA levels among different clinical manifestations(all P<0.05).Pathological grouping indicated that there were significant differences in sex ratio,onset age,course of disease,family history,Addis count of urinary red cells,24-hour proteinuria,blood urea nitrogen,serum creatinine,serum albumin,serum IgA and C3 levels among different pathological groups(all P<0.05).The BP neural network prediction model was then constructed based on the above indicators,and the accuracy of the prediction model was measured to be 61.19%by using the leave one out method.Conclusions By comparing the differences of various indicators under different clinical manifestations and pathological types,a BP neural network prediction model for renal pathology in children with hematuria as the main manifestation is established.The model can accurately predict renal pathology with the help of related indicators,and provides a basis for determining the time of kidney biopsy.