小儿常见发热出疹性疾病智能诊断模型研究

Development of an intelligent model for diagnosis of common rash and fever illness in children

目的探索实用于社区医生和家庭成员使用的小儿常见发热出疹性疾病智能诊断方法。方法收集2005年1月至2010年11月第三军医大学西南医院儿科及感染科248例小儿发热出疹性疾病住院患者的临床资料,其中男性133例,女性115例,平均年龄4.56岁。病种包括麻疹、幼儿急疹、水痘、手足口病、猩红热、风疹和药疹等。整理并描述发热、皮疹、主要伴随症状、血常规及流行病学相关数据特征,进行主成分分析(PCA);以反向传播神经网络(BPNN)为技术平台,构建智能诊断模型,进一步通过前瞻和回顾数据验证模型的准确性。结果经PCA处理后,31个临床及流行病学特征指标被综合成13个主因子;BPNN模型的输入、隐层和输出神经元分别为13、9、7;模型对小儿发热出疹性疾病回顾性诊断平均准确率达到99.53%,预测诊断平均准确率达到92.86%。结论以临床样本为依据建立的BPNN诊断模型可准确诊断常见小儿发热出疹性疾病,有明显的应用前景。

Objective To explore an intelligent model for diagnosis of common rash and fever illness（RFIs） in children for medical staff in rural area and family members.Methods Clinical data of 248 RFIs cases（including 133 males and 115 females with an average age of 4.56） were collected from inpatients in the Southwest Hospital of The Third Military Medical University from Jan 2005 to Nov 2010.Diseases comprised of measles,exanthem subitum,chicken pox,hand-foot-and-mouth disease,scarlet fever,rubella and exanthema.Features of fever,rash,main concomitant symptoms,blood routine and epidemiological data were organized and described,and principal component analysis（PCA） was carried out.PCA combined with back-propagation neural network（BPNN） was used to set up an intelligent diagnosis model for children with common RFIs.The accuracy of the model was further confirmed based on prospective and retrospective analysis.Results Thirty-one clinical and epidemiological variables were integrated into 13 principle factors through PCA.These factors were then input to set up a BPNN with a 13-9-7 structure.When the model was used for children with RFIs,the average accuracy rate of retrospective diagnosis reached 99.53%,and the average accuracy rate of predictive diagnosis was 92.86%.Conclusion BPNN diagnosis model on the basis of clinical samples can be applied for an accurate diagnosis of common RFIs in children,and has an significant application prospect.