目的

采用决策树的方法构建尿液镜检规则，以提高标本筛选的准确率。

方法

使用UF-1000i、Urisys 2400尿液干化学分析仪（以下简称Urisys 2400）及DiaSys RS 2003尿沉渣工作站（以下简称RS 2003）分析2000例尿液样本。根据其检查结果，定义“阳性样本”和“阴性样本”。在此基础上，构建训练集及测试集。并采用决策树的方法进行建模预测，以获得最优的镜检规则。

结果

基于决策树方法构建的尿液镜检规则，镜检率能控制在25.0%左右，敏感度为85.0%，特异度为96.4%。与已有镜检规则相比，该镜检规则的镜检率较低。

结论

决策树方法可以用于尿液分析镜检规则的建立，该方法对现有的镜检规则是有益的补充。

Objective

To get better accuracy, microscopic review rules of urine sediments were built based on a decision tree approach.

Methods

A total of 2000 urine samples were examined using UF-1000i, Urisys-2400, and RS 2003 urine sediment workstation, respectively. Positive and negative samples were defined, and both the training and testing sets were set up. Finally, a decision tree approach was employed to construct classifiers for screening urine samples.

Results

Using the decision tree method, we obtained a sensitivity of 85.0%, a specificity of 96.4% and a total review rate of 25.0% on the testing set, showing the acceptable sensitivity and lower total review rate comparing with the existing method.

Conclusion

An algorithm based on the decision tree for building review criteria can be available, which is valuable and a supplement for other microscopic review rules.