文章摘要

文章信息/Info

Title:: Prediction of disease-free survival in patients with hepatocellular carcinoma based on imbalance classification

作者:: 杨日东1; 李琳2; 陈秋源1; 华赟鹏3; 周毅1△; 1.中山大学中山医学院，广州 510080；2.新疆医科大学公共卫生学院，乌鲁木齐 830011；3.中山大学附属第一医院肝胆外科，广州 510080

Keywords:: Hepatocellular carcinoma; Disease-free survival; Class imbalance; Boosting tree; Sampling technique

摘要:: 为了预测原发性肝癌患者的术后无病生存期能否达到5年。针对患者临床数据类不平衡的特性，修改传统提升算法中样本权重和基分类器权重的更新方式，并结合决策树，提出一种改进的提升树算法。在患者的临床数据集上，将改进的提升树与传统提升树、结合多种采样技术的决策树作对比。发现改进提升树的F值、G均值和AUC等指标都取得最优值。在所研究的方法中，改进的提升树在原发性肝癌患者的5年无病生存期预测中性能最佳，可为同类研究提供方法学参考。

Abstract:: To predict whether the disease-free survival time of patients with hepatocellular carcinoma can over 5 years.Aiming at the imbalance of clinical data, we modified the update method of sample weight and base classifier weight in traditional boosting algorithm. Combined with decision tree, an improved boosting tree algorithm was proposed.Compared with the traditional boosting tree and the decision tree combined with multiple sampling techniques on the clinical data sets, the F measure, G mean and AUC of the improved boosting tree were all optimal. Among the methods studied in this paper, the improved boosting tree has a best performance on predicting the 5-year disease-free survival of patients with hepatocellular carcinoma. It can provide a methodological reference for similar research.