广义回归神经网络模型诊断酒精性肝病的初步研究  

A diagnostic model for alcoholic liver disease based on a generalized regression neural network

作　　者：李耐萍[1] 蒋永芳[1] 

机构地区：[1]中南大学湘雅二医院感染科,长沙410007

出　　处：《中华肝脏病杂志》2013年第4期304-307,共4页Chinese Journal of Hepatology

摘　　要：目的探讨广义回归神经网络（GRNN）模型在酒精性肝病（ALD）诊断中的可行性。方法建立ALD诊断的GRNN模型，其输入层7个指标分别为γ-谷酰氨转移酶、总胆汁酸、碱陛磷酸酶、总胆红素、ALT、AST及AST／ALT值，输出层3个指标分别为酒精性肝硬化肝功能失代偿期、酒精性肝硬化肝功能代偿期和酒精性肝炎。收集湘雅二医院确诊并分类的135例患者，选取其中120例为模型训练样本，另外15例作为待诊断样本；同时，收集文献发表的40例酒精性肝炎患者的临床数据，以其中34例为模型训练样本，6例作为待诊断样本。运用GRNN方法对其进行运算与诊断，检验GRNN模型的诊断结果与临床诊断结果的符合率。结果用建立的GRNN模型分别对120例和34例模型训练样本进行回判诊断，其诊断结果与临床诊断结果的符合率分别为100．00％和94．1砒；对15例和6例待诊断样本进行诊断，符合率均为100．00%。结论GRNN模型作为临床诊断多指标综合化的方法，对ALD的诊断有较高的准确率。Objective To study the feasibility and rationale of using a generalized regression neural network model integrated with multiple disease indicators for diagnosing alcoholic liver disease （ALD）. Methods ALD indicators were identified by reviewing the clinical testing results of 40 ALD patients from the literature and 135 patients from the Second Xiangya Hospital of Central South University, who were also classified by physician experts upon clinical consultation. Seven indicators were selected as diagnosis indexes and applied to a general regression neural network diagnostic model. Thirty-four of the reported patients and 120 of the clinical patients were selected for use as training samples to establish the indicator recognition pattern for the model, and the remaining six and 15 patients from the two respective groups were selected for use as testing samples to determine the model＇s diagnostic ability. Results The model provided a correct diagnosis of ALD sub-classification for 94.1% （32/34） of the reported patients and 100% （120/120） of the clinical patients in the training set. The correct diagnosis rates achieved with the training sets were 100% for both the reported patient group （6/6） and the clinical patient group （15/15）, indicating that the results of the diagnostic model were in good agreement with the ALD classifications generated by the clinical expert consultations. Conclusion The general regression neural network model based on multiple indicators of ALD is capable of providing accurate and comprehensive diagnosis of ALD and may be feasible for clinical applications.

关 键 词：肝疾病 酒精性 诊断 肝功能试验 神经网络(计算机) 

分 类 号：R575.5[医药卫生—消化系统]