The Classification And Prediction Model Of HBV Reactivation In Liver Cancer Patients Based On Feature Selection

Hepatitis B virus(HBV)reactivation is a common complication after precise radiotherapy in patients with primary liver cancer(PLC).In this paper,we established the classification prediction model of HBV reactivation in patients with PLC after precise radiotherapy.Using the classification prediction model we can provide the reference to the doctor to prevent diseases and reduce the incidence of the disease.The clinical data of 90 HBV-related PLC patients were collected from Shandong cancer hospital.Each sample texts involves twenty-eight characteristics.Because not all of the 28 features are dangerous factors and there are data redundancies,we need to use the feature selection algorithm to select several risk factors from the 28 features,and then establish classification prediction models.In this paper,we adopt two models to build classification prediction model for HBV reactivation.The first model is to use sequential selection and principal component analysis to build classification prediction model for HBV reactivation.The second scheme is to use one-dimensional continuous wavelet and random forest to build classification prediction model for HBV reactivation.The experimental results showed that we use sequential forward selection find that times of radiotherapy,HBV DNA levels,outer margin of radiotherapy,AFP and split method were the best risk factors of HBV reactivation and the highest accuracy of the key features reached to 84.04%.Among them times of radiotherapy,AFP and split method are newly proposed risk factor.We use sequential backward selection select the 5 risk factors HBV DNA levels,KPS score,outer margin of radiotherapy,equivalent biometric and tumor stage TNM,and the highest accuracy reached to 87.31%.Equivalent biometric is a new risk factor.The combination of the above two methods selected risk factors was analyzed by PCA,and eight risk factors were found to be the risk factors affecting the reactivation of HBV,and there was no redundant information.In the second model,we find out that tumor stage TNM,outer margin of radiotherapy,HBV DNA levels,V10 and V20 were the risk factors by using random forest.We established random forest and Bayesian classification prediction models and the highest accuracy of the key features reached to 84.66%.To eliminate the noise in the data,we use one-dimensional continuous wavelet transformation to process the original data and then use random forest to select the 5 risk factors equivalent biometric,HBV DNA levels,KPS score,portal venous emboli and GTV volume,the accuracy of random forest classification prediction model reached to 82.11%.The classification prediction models of HBV reactivation proposed in this paper can be used to solve the problem of HBV reactivation classification and prediction,breaking the limit of using traditional medical methods to find out the risk factors and realizing the close combination of the computer field and the medical field.