基于人工神经网络的前列腺癌诊断模型对前列腺癌的诊断价值研究  

Artificial Neural Network-based Diagnostic Model for Prostatic Cancer

作　　者：宋敏[1] 王开正[1] 杭永伦[1] 李光荣[1] 田刚[1] 刘靳波[1] 

机构地区：[1]泸州医学院附属医院检验科,四川省泸州市646000

出　　处：《中国全科医学》2012年第35期4061-4063,共3页Chinese General Practice

基　　金：四川省卫生厅科研课题([2010]493号100258)

摘　　要：目的结合前列腺肿瘤标志物检验组套和患者临床信息进行数据挖掘,建立基于人工神经网络(ANN)的前列腺癌诊断模型,为前列腺癌的临床诊断和治疗提供客观的参考信息。方法通过实验信息系统与医院信息管理系统的数据信息平台检索并统计2010年1月—2011年7月我院前列腺肿瘤标志物检验组套病例365例,其中前列腺癌组60例,非前列腺癌组305例。采用受试者工作特征(ROC)曲线下面积法筛选出有价值的指标,用244例样本(前列腺癌组40例,非前列腺癌组204例)建立ANN模型,并用121例样本(前列腺癌组20例,非前列腺癌组101例)盲法测试和评估此模型。结果纳入分析的指标有年龄、甲胎蛋白(AFP)、癌胚抗原(CEA)、总前列腺特异抗原(tPSA)和结合前列腺特异抗原(cPSA),各指标的曲线下面积分别为0.623、0.517、0.499、0.907和0.913,其中年龄、tPSA和cPSA与前列腺癌的发病有相关性(P<0.05);经方差分析前列腺癌组的年龄、tPSA和cPSA与非前列腺癌组比较,差异均有统计学意义(P<0.05)。建立的模型对训练样本预测的特异度为93.63%,敏感度为82.50%;此模型对121例测试样本预测的特异度为93.07%,敏感度为80.00%。结论数据挖掘技术能够提炼出高效的诊治信息,基于ANN的前列腺癌诊断模型对前列腺癌的早期诊断具有一定价值。Objective To establish diagnostic model for prostatic carcinoma based on artificial neural network （ANN） by combining the serum markers of prostatic carcinoma and clinical information in order to provide referenees for clinical diagnosis and treatment of prostatic carcinoma. Methods Based on experiment information system and hospital information system, 365 patients whose serum markers of prostatic carcinoma were tested and collected from January 2010 to July 2011 were retrieved and they were divided into prostatic carcinoma group （ 60 cases） and non - prostatic carcinoma group （ 305 cases）. The indicators were evaluated with the method of area under the ROC curves, and 244 cases （40 cases from prostatic carcinoma group and 204 cases from non -prostatic carcinoma group） were used to built the diagnostic model with artificial neural network and 121 samples （20 samples from prostatic carcinoma group and 101 samples from non- prostatic carcinoma group） were used to assess this mod- el. Results Age, AFP, CEA, tPSA and cPSA were involved into the analysis. The areas under the curve of the indicators were 0. 623, 0. 517, 0. 499, 0. 907 and 0. 913, respectively. The incidence of prostatic carcinoma were related to age, tPSA and cPSA （ P 〈 0.05 ） and the three indicators showed statistically significant differences between prostatic carcinoma group and non -prostatic carcinoma group by analysis of variance （P 〈 0.05）. The specificity and sensitivity of this model were 93.63% and 82. 50% for the exercise sample and 93.07% and 80. 00% for the test sample. Conclusion The technology of data mining can extract effective information of diagnosis and treatment. The diagnostic model for prostatic carcinoma which was based on arti- ficial neural network may be a valuable clinical tool for early diagnosis of prostatic carcinoma.

关 键 词：前列腺肿瘤 肿瘤标记 生物学 神经网络(计算机) ROC曲线 诊断 

分 类 号：R737.25[医药卫生—肿瘤]