To Evaluate The Sensitivity Of Chemoradiotherapy In Non-small Cell Lung Cancer By Radiomics Based On Enhanced CT Images Before Chemoradiotherapy

Objective: To evaluate the sensitivity of chemoradiotherapy in non-small cell lung cancer(NSCLC)by radiomics based on enhanced CT images before chemoradiotherapy.Methods: 125 patients(236 lesions)with NSCLC confirmed by pathology in the General Hospital of the Northern Theater Command from November 2010 to December 2019 were retrospectively analyzed.The patients were randomly divided into the validation group and the training group according to the ratio of 1:4.Response Evaluation Criteria in Solid Tumors(RECIST 1.1)served as the gold standard for sensitivity evaluation of chemoradiotherapy.Dr.Wise multimodal scientific research platform software was used to extract the radiomics features,screen the features and establish the radiomics labels.The clinical data of patients and radiomics labels were included.Logistic regression(LR),support vector machine(SVM),decision tree(DT),random forest(RF)and linear support vector machine(LSVM)were used to establish the models.The accuracy,sensitivity,specificity and Area under receiver operating characteristic(ROC)curve were used to compare the performance of the five classifiers,and the model was optimized and verified.Results: There was a positive correlation between radiomics label and the assessment of chemoradiotherapy sensitivity(P<0.001).The five classifier methods were modeled in five-fold cross validation.In the LR classifier model training group,the accuracy,sensitivity,specificity were 0.66,0.72 and 0.62,and the area under ROC curve was0.76(95% CI value 0.709-0.812);validation group accuracy,sensitivity and specificity was 0.44,0.62,0.53,and and the area under ROC curve was 0.44(95% CI value0.591-0.709).In the SVM classifier model training group,the accuracy,sensitivity,specificity were 0.82,0.74 and 0.87,and the area under ROC curve was 0.90(95% CI value 0.856-0.935);validation group accuracy,sensitivity and specificity was 0.70,0.53,0.82,and and the area under ROC curve was 0.76(95% CI value 0.700-0.824).In the DT classifier model training group,the accuracy,sensitivity,specificity were 1.0,1.0 and 1.0,and the area under ROC curve was 1.0(95% CI value 1.0-1.0);validation group accuracy,sensitivity and specificity was 0.47,0.52,0.66,and the area under ROC curve was 0.66(95% CI value 0.604-0.722).In the RF classifier model training group,the accuracy,sensitivity,specificity were 1.0,1.0 and 1.0,and the area under ROC curve was 1.0(95%CI value 1.0-1.0);validation group accuracy,sensitivity and specificity was 0.54,0.23,0.88,and and the area under ROC curve was 0.62(95% CI value 0.562-0.682).In the LSVM classifier model training group,the accuracy,sensitivity,specificity were 0.98,0.91 and 0.99,and the area under ROC curve was 0.99(95% CI value 0.995-1.0);validation group accuracy,sensitivity and specificity was 0.56,0.46,0.78,and the area under ROC curve was 0.68(95% CI value 0.624-0.738).Conclusion: The area under ROC curve of SVM classifier is the highest,so the SVM classifier model is the best.The model established by SVM classifier method screened out 21 meaningful radiomics features.The sensitivity of RECIST and SVM models are similar in predicting chemoradiotherapy of NSCLC.The SVM classifier method model based on the radiomics label obtained from enhanced CT images of lung cancer before chemoradiotherapy has certain application value for the sensitivity evaluation of chemoradiotherapy in NSCLC.