中文摘要：

目的探讨基于人工神经网络（ANN）判别模式的肿瘤标志物联合检测对肝癌辅助诊断的价值。方法 2009年3月-2010年3月解放军159中心医院和郑州大学第一附属医院门诊、住院患者及体检者140例,分别纳入肝癌组（n=50）、肝良性病变组（n=40）和正常人组（n=50）。采用化学发光免疫检测试剂盒测定血清中甲胎蛋白（AFP）、癌胚抗原（CEA）、糖类抗原125（CA125）的含量,采用可见光分光光度法测定血清唾液酸（SA）水平,采用偶氮砷Ⅲ终点法测定血清中的钙（Ca）含量,以此5种肿瘤标志物作为判别变量,运用Fisher判别分析法及反向误差传播ANN技术,建立肝癌智能化辅助诊断模型。结果本研究建立的Fisher判别分析模型对3组样本判别的灵敏度为46.1%,特异度为98.9%,准确度为79.3%,总的阳性预测值为95.8%,总的阴性预测值为76.7%;而ANN模型对3组样本判别的灵敏度为96.0%,特异度为98.9%,准确度为94.3%,阳性预测值为98.0%,阴性预测值为97.8%。结论多个肿瘤标志物联合ANN技术建立的肝癌诊断模型对肝癌的预测效果优于传统的Fisher判别分析方法,更适用于临床数据的判别分析。

英文摘要：

Objective To explore the value of determination of combined tumor markers based on artificial neural network （ANN） discrimination model in facilitating the diagnosis of hepatic carcinoma. Methods Serum samples were collected from three groups of subjects, including 50 cases of liver cancer, 40 cases of benign liver disease, and 50 normal controls. The levels of serum alpha fetoprotein （AFP）, carbohydrate antigen 125 （CA125） and carcino-embryonic antigen （CEA） were determined by chemiluminescence immunoassay. The level of serum sialic acid （SA） was determined by spectrophotometry, the content of calcium in serum was measured by calcium assay kit （Azo-end method of arsenic III ）. Based on the five tumor markers mentioned above as discrimination variables, Fisher discrimination and ANN were applied to set up the intelligent auxiliary diagnostic model. Results By applying the Fisher discrimination model established in present work, the diagnostic sensitivity of liver cancer was 46.1%, the specificity was 98.9%, the accurate rate was 79.3%, the positive predictive value was 95.8%, and the negative predictive value was 76.7% for the three groups. With the application of ANN discrimination model, the diagnostic sensitivity of liver cancer was raised to 96.0%, the specificity 98.9%, the accuracy 94.3%, the positive predictive value 98.0%, and the negative predictive value was 97.8%. Conclusion The diagnostic model based on ANN combined with 5 tumor markers is superior in diagnostic acuity to traditional Fisher discrimination analysis, thus more suitable for clinical data analysis.