A Feature Level Fusion Research Based On Rough Sets For Lung Tumor PET/CT Image

Background Lung cancer poses a great threat to human health.Nowadays,medical imaging provides advanced clinical examination techniques for patients with lung cancer,among them,PET/CT has the function of PET imaging and the anatomical function of CT,both of which are complementary to each other,but the mass of medical imaging increased the reading burden of doctors,so rough sets based computer aided diagnosis of lung tumor can provide quantitative analysis to the doctor and provide reference with good consistency,reduce the workload and improve the diagnostic effect of doctor diagnosis.Objectives The CT,PET and PET/CT images of lung tumor as the research objects,ensemble SVM and variable precision rough set model were proposed to realize the computer-aided diagnosis of lung tumors,reduce the workload of medical diagnosis,improve the efficiency of reading.Methods On the basis of rough set model structure,the ensemble SVM and variable precision rough set two kinds of models were proposed for lung tumor image recognition.The ensemble SVM was used to identify the features of lung tumor PET/CT three modal images,the variable precision rough set model was used to identify the features of PET/CT images,and the accuracy,sensitivity,specificity and time are taken as the performance index.Results Aiming at the ensemble SVM based the computer aided diagnosis research of lung cancer PET/CT images,five experiments were performed: constructing individual classifiers in CT feature space,constructing individual classifiers in PET feature space,constructing individual classifiers in PET/ CT feature space,computer aided diagnosis based on ensemble learning,disturb the 10%data of CT sample space and construct individual classifier.The experimental results show that the ensemble SVM is superior to the single SVM in classification accuracy,and can improve the fault tolerance of the classifier.Aiming at feature level fusion based on genetic algorithm and variable precision rough set model,three kinds of experiments were performed: study on different weight coefficients based on the same classification error rate,study on different classification error rate based on the same weight coefficients,study on increasing the weight coefficient based on the same classification error rate.At the same time,the feature level fusion based on genetic algorithm and rough set model was compared.The experimental results show that the proposed method is superior to the traditional methods in the recognition of lung tumors.At the same time,selecting the appropriate parameter combination can get better recognition effect.