基于随机森林的慢性丙型肝炎纤维化和活动度分析

为对慢性丙型肝炎病情相关的肝纤维化阶段和炎症活动度进行预测,提出一种基于两阶段随机森林的自动分级方法。首先,在训练病例集上进行第一阶段随机森林模型学习,获取各个血清学指标的特征重要度,以衡量这些指标与肝炎纤维化阶段和炎症活动度之间的相关程度;其次,选择特征重要度大于阈值的血清学指标,可作为下一步进行分类的特征;最后,在选出的显著性特征上进行第二阶段的随机森林模型训练,对慢性丙型肝炎的肝纤维化程度和炎症活动度进行自动分级。通过对123例慢性丙型肝炎的血清学指标进行分析,得到纤维化阶段、纤维化S₄阶段和炎症活动度的分类正确率分别为68.29%、100%和73.17%,得到与慢性丙型肝炎纤维化分期和活动度分级密切相关的重要血清指标为总胆固醇、高密度脂蛋白胆固醇、谷丙转氨酶、天门冬氨酸转氨酶等。实验结果表明,采用的检验指标获取成本低、计算量低,能达到较好的分级准确度,有助于慢性丙型肝炎诊断。

Fibrosis and Inflammatory Activity Analysis of Chronic Hepatitis C Based on Random Forest

In order to predict the fibrosis stage and inflammatory activity grade of chronic hepatitis C, an auto-grading system based on two-stage random forest was proposed in this paper. Firstly, the feature importance of each serological index was obtained by learning the first stage random forest to evaluate its relevance to fibrosis stage and inflammatory activity grade. Secondly, the serological indices whose feature importance were above the predetermined threshold were chosen for the next classification step. Finally, the second stage random forest based on the chosen features was employed for determining the fibrosis stage and inflammatory activity grade. The proposed method has been tested on 123 clinical data of chronic hepatitis C based on serological indexes. Experimental results showed that the classification accuracy of fibrosis stage, fibrosis stage S4 and inflammatory activity grade are 68.29%, 100% and 73.17%. At last the most important serological indexes related to the fibrosis stage and inflammatory activity level ofchronic hepatitis C were determined as total cholesterol, HDL, ALT and AST. Experimental results demonstrated that the proposed method has the advantages of high recognition accuracy and low cost to get examination results and perform calculations, which makes it helpful for clinical diagnosis of chronic hepatitis C.