A Study on decision-support system of clinical laboratory tests.

对肝病临床实验室诊断决策支持系统进行了研究。该系统由两个步骤组成。一个是利用人工神经网络的步骤，另一个是利用医学规则库的步骤。在前者中，设计了一个反向传播神经网络来诊断七种慢性肝病：慢性非活动性肝炎、慢性活动性肝炎、肝硬化、肝细胞癌、脂肪肝、酒精性肝病和正常。将神经网络的输出输入到第二步进行最终诊断。网络的输入数据是来自187例病例的26项肝脏实验室数据。该系统诊断符合率为75.4%（141/187），高于5位病理学家的诊断符合率（63.0%）。该系统作为慢性肝病的决策支持系统似乎是有效的。应用模糊理论对肝病进行诊断时，敏感性高，特异性低，然后从30组实验数据中，尝试用人工神经网络和决策树逻辑对炎症性疾病、肌肉或心肌疾病、贫血、恶性肿瘤、肾泌尿系统疾病、肝胆疾病、糖尿病、胃肠道疾病、骨病和高脂血症10个疾病组进行分类。对10个疾病组进行了自动分类，诊断正确率为80.8%，该系统将有助于从基础实验室数据推断疾病组。总之，神经网络是非常强大的区分群体，显示重叠的参数限制。

A study on decision-support system for liver disease using clinical laboratory tests was performed. The system was composed of two steps. One is a step to use an artificial neural network and the other is a step to use medical rule base, In the former, a back propagation neural network was designed to diagnose seven categories of chronic liver diseases : chronic inactive hepatitis, chronic active hepatitis, liver cirrhosis, hepatocellular carcinoma, fatty liver, alcoholic liver desease, and normal. The output of a neural network were input to the second step for a final diagnosis. Input data to the network were 26 items of hepatic laboratory data from 187 cases. The diagnostic accuracy of the system was 75.4% (141 of 187 cases) which was higher than the diagnostic sccuracy (63.0%) of 5 hapatologists. Yhis system was seemed to be effective as a decision-support system for chronic liver disease. When Fuzzy theory was used to diagnose liver disease, high sensitivity was obtained with low specificity.Next, we tried categorization of ten disease groups of inflammatory disease, muscular or myocardial disease, anemia, malignant tumor, reno-urinary disease, hepatobiliary disease, diabetes mellitus, gastrointestinal disease, bone disease and hyperlipidemia from 30 sets of leboratory data using artificial neural network and dicision-tree logic. Ten-disease groups were automatically classified with 80.8% of diagnostic accuracy.This system will be helpful to infer the disease groups from basic laboratory data. In conclusion, neural network in very powerful to differentiate groups which shows overlapped limits of parameters.