原发性肝癌放疗后乙肝病毒再激活预测模型的特征降维分析

发布时间：2018-06-12 10:55

本文选题：乙肝病毒(HBV)再激活 + 顺序特征选择　；参考：《中国生物医学工程学报》2017年06期

【摘要】：建立对原发性肝癌(PLC)患者精确放疗后乙肝病毒(HBV)再激活的分类预测模型可以提前进行预防治疗,减少发病率。采用山东省肿瘤医院的90例原发性肝癌患者放疗后的临床指标作为研究数据。每例患者的检查记录都包括:性别、年龄、KPS评分、AFP、HBV DNA水平、TNM等30个特征。提出用顺序特征选择进行关键特征的选取,将选取的特征组成新的特征子集,并建立贝叶斯分类预测模型。顺序后向选择发现KPS评分、HBV DNA水平、外放边界、TNM、全肝最大剂量是乙肝病毒再激活的危险因素,建立贝叶斯分类模型,结果采用3折交叉验证,预测精度达到85.75%。顺序前向选择发现性别、KPS评分、HBV DNA水平、HBe Ag、外放边界两分类编码是乙肝病毒再激活的危险因素,5折交叉验证下的贝叶斯分类预测精度达到84.06%。实验结果表明,贝叶斯分类器可以很好地用于乙肝病毒再激活的研究,特征选择后的关键特征具有更优越的分类性能。
[Abstract]:The classification prediction model of HBV reactivation after accurate radiotherapy in patients with primary liver cancer (PLC) can be used to pretreat the patients and reduce the incidence of hepatitis B virus (HBV) reactivation in advance. The clinical data of 90 patients with primary liver cancer after radiotherapy in Shandong Cancer Hospital were used as the data. The examination records of each patient included 30 characteristics, such as gender, age, KPS score, HBV DNA level and TNM. The selection of key features by sequential feature selection is proposed. The selected features are formed into a new feature subset and a Bayesian classification prediction model is established. The KPS score and HBV DNA level were found in the sequential backward selection. The maximum dose of whole liver was the risk factor of HBV reactivation. The Bayesian classification model was established, and the prediction accuracy was 85.75% by 3 fold cross validation. In the sequence forward selection, it was found that gender KPS score and HBV DNA level were higher than HBe Ag.The prediction accuracy of Bayesian classification was 84.06 under the cross validation of HBV reactivation risk factor (5%). The experimental results show that the Bayesian classifier can be used to study the reactivation of hepatitis B virus, and the key features after feature selection have better classification performance.
【作者单位】：齐鲁工业大学信息学院;山东省肿瘤医院肿瘤放疗科;
【基金】：国家自然科学基金(81402538,61375013) 山东省自然科学基金(ZR2013FM020)
【分类号】：R512.62;R735.7

【相似文献】