基于多个神经网络模型组合诊断乙型肝炎肝纤维化的研究  

Study on multiple neural network model for diagnosis of hepatitis B hepatic fibrosis

作　　者：李波[1] 孙志强[1] 李筱涵[1] 李小溪[1] 毛远丽[1] 

机构地区：[1]中国人民解放军第三〇二医院临床检验中心,北京100039

出　　处：《国际检验医学杂志》2012年第20期2457-2459,共3页International Journal of Laboratory Medicine

基　　金：军队"十一五"科技攻关课题(06G143)

摘　　要：目的为提高神经网络模型诊断肝纤维化的灵敏度和特异度,联合多个网络模型进行诊断。方法留取84例具有明确病理诊断的患者样本,根据参考文献选择适合肝纤维化诊断的血清学及血液学指标,建立3个不同的神经网络诊断模型,通过分层随机方法分成训练组和验证组,以病理诊断结果作为金标准,计算模型判断的准确率、灵敏度、特异度等指标。结果三个模型对乙型肝炎肝纤维化诊断的准确率为分别为74%、76%、68%,敏感度分别为62%、84%、50%,特异度为79%、70.8%、87.5%。三个模型结果进行联合分析诊断的准确率、灵敏度与特异度分别为82%、76.9%、87.5%,结果优于单个模型诊断价值。结论联合多个神经网络模型较单一模型对乙型肝炎导致的肝纤维化诊断具有较高的敏感度及特异度。Objeetive To improve the sensitivity and specificity of the neural network model in diagnosis of liver fibrosis,a com- bined model with multiple network models for liver fibrosis diagnosis was established. Methods A total of 84 patient samples with biopsy were collected. According to some references, we selected several serum or blood markers to established 3 different models. The 84 cases were divided into training subset and validation subset by stratified sampling. Using biopsy results as gold standard, accuracy,sensitivity and specificity of the model were assessed. Results 3 models＇ accuracy were 74%,76 % and 68 %, Sensitivity were 62% ,84%, 500%, and specificity were 79%, 70. 8%, 87. 5% respectively. Accuracy, sensitivity and specificity of combined model were 82 %,76.9 % and 87.5 % , which was better than the any single model. Conclusion Combining more than one models had better sensitivity and specificity than any of individual model.

关 键 词：人工神经网络 肝纤维化 活检 

分 类 号：R575.2[医药卫生—消化系统]