基于连续小波和随机森林的原发性肝癌放疗后乙肝病毒再激活的分类预测

Classification of Hepatitis B Virus reactivation in patients with Primary Liver Cancer after precise radiotherapy based on Continuous Wavelet Transform and random forest

原发性肝癌(PLC)患者精确放疗后乙肝病毒(HBV)再激活是一种常见的并发症,影响患者预后,危及患者的生命。通过连续小波进行去噪,然后再通过随机森林对特征重要性的排序找出引发乙肝病毒再激活的危险因素,给医生提供参考,进而提前进行预防治疗,降低病毒再激活的发病率。首先对原始数据集进行连续小波变换,之后使用随机森林进行关键特征的选取,将随机森林模型下的特征按照重要性进行排序,选取重要性最高的5个特征组成关键特征子集,然后将新的特征子集用随机森林分类器进行分类预测。实验结果表明随机森林选取HBV DNA水平、TNM肿瘤分期、V10、V20、外放边界这5个关键特征作为致使乙肝病毒再激活的危险因素组合时,进行小波变换后,3折交叉验证下预测精度最高达到82.96%。本次研究表明,小波变换后可以有效地降噪,随机森林可以通过评估变量的重要性,选出关键特征,很好地用于解决乙肝病毒再激活分类预测问题。

Hepatitis B Virus（ HBV） reactivation is a common complication in patients with Primary Liver Cancer（ PLC）after the precise radiotherapy. It affects the prognosis of patients and endangers the patient lives. This paper aims to find out the risk factors of HBV reactivation by random forest. It would provide the reference for the doctor and reduce the incidence of the disease. In this paper,firstly,use Continuous Wavelet Transform（ CWT） to deal with the raw data; after that,use random forest method to select key features. All the features are sorted according to the importance. Based on the aboved,4and 5 most important features are selected which would be combined into a brand new feature subset and then random forest classification prediction model is established. The experimental results show that such 5 risk factors as HBV DNA level,TNM tumor staging,V10,V20,outer margin of radiotherapy are best combination of HBV reactivation. After the Wavelet Transform,the best classification accuracy of random forest can be reached to 82. 96% using 3 fold cross validation. The research shows that Wavelet Transform can effectively denoise and the random forest can be used to evaluate the importance of variables and select the key features. And also,it is a better method to solve the classification prediction problem of HBV reactivation.