The Study Of Radiomics Based On ¹⁸F-FDG PET/CT In Patients With Non-Small Cell Lung Cancer

Part 1Objective To distinguish non-small cell lung cancer(NSCLC)from pulmonary tuberculosis(PTB)presenting as nodules or masses using 18F-FDG PET-based radiomic features.Methods A retrospective analysis was performed in 231 patients with PTB and NSCLC who underwent ¹⁸F-FDG PET/CT scans in Tongji hospital of Huazhong University of Science and Technology.The patients with NSCLC were divided into a training and validation set at a ratio of 7:3.Radiomic features were extracted from the PET images using the python package.The minimum Redundancy Maximum Relevance feature selection(mRMR)algorithm and Least Absolute Shrinkage and Selection Operator(LASSO)were employed to select informative and non-redundant features,and a radiomics signature score(Rad-score)was developed.Differences between groups were tested using the Mann-Whitney U test.Multivariate logistic regression was applied to select the important factors.We constructed a combined model based on the clinical variable and radiomics signature,and compared the predictive performance of models using receiver operating curves(ROC).Results Six radiomics features were selected to build the Rad-score.The Rad-score showed a significant ability to discriminate between different histological types in the two sets(training set:z=-8.53,P<0.001;validation set:z=-5.11,p<0.001),with area under the ROC curve(AUC)equal to 0.967(95%CI,0.926-0.989)in the training set,and 0.915(95%CI,0.823-0.969)in the validation set,compared with AUC=0.774,0.729 for the clinical variable.When clinical variables and radiomics signature were combined,the complex model showed better performance in the classification of histological types,with the AUC increased to 0.981(95%CI,0.9470.996)in the training set and 0.941(95%CI,0.857-0.984)in the validation set,significantly higher than that of SUVmax(training set:z = 5.178,p<0.001;validation set:z=2.533,p=0.011).Conclusion ¹⁸F-FDG PET/CT-based radiomics features showed good performance in distinguishing PTB from NSCLC,which would help to improve the diagnostic accuracy of pulmonary lesions.Part 2Objective To distinguish lung adenocarcinoma(ADC)from squamous cell carcinoma(SCC)using 18F-fluorodeoxyglucose(FDG)PET/CT based radiomics features.Methods A retrospective analysis was performed in 182 patients with non-small cell lung cancer(NSCLC)who underwent ¹⁸F-FDG PET/CT scans in Tongji hospital of Huazhong University of Science and Technology.All patients had been diagnosed pathologically with lung ADC or SCC.The patients were divided into a training set and a validation set at a ratio of 1:1.Radiomic features were extracted from the PET and CT images of segmented tumors using the python package.The minimum Redundancy Maximum Relevance feature selection algorithm and Least Absolute Shrinkage and Selection Operator were employed to select informative and nonredundant features,and a radiomics signature score(Rad-score)was developed.Differences between different groups were tested using Mann-Whitney U test.Multivariate logistic Regression was applied to select the important factors.We constructed a combined model based on the clinical variable and radiomics signature,and compared the predictive performance of models using receiver operating curves and Delong test.Results Five radiomic features(gldm_LargeDependenceLowGrayLevelEmphasis,glszm_GrayLevelVariance,gldm_LargeDependenceHighGrayLevelEmphasis,glcm_SumEntropy,firstorder_Maximum)were selected to build the Rad-score.The Rad-score showed a significant ability to discriminate between different histological subtypes in the two sets[training set:-1.40(-2.16-0.36)vs-0.46(-1.30?0.89),z=-5.797,p<0.001;validation set:-1.40(-2.21--0.43)vs-0.53(-1.69-0.95),z=-5.314,p<0.001],with area under the ROC curve(AUC)equal to 0.895(95%CI,0.813-0.950)in the training set and 0.850(95%CI,0.760-0.916)in the validation set,compared with 0.721 and 0.726 for the clinical variable.When clinical variables and radiomics signature were combined,the complex model showed better performance in the classification of histological subtypes,with the AUC increased to 0.912(95%CI,0.834-0.961)in the training set and 0.875(95%CI,0.789-0.935)in the validation set.Conclusion Individualized diagnosis model incorporating with smoking and radiomics signature could help differentiate histological subtypes in a non-invasive,repeatable modality.Part 3Objective Tumor staging is one of the most important factors in treatment decisions and prognosis.N2 lymph node status determines whether surgical treatment or neoadjuvant chemotherapy is the best intervention,and preoperative risk stratification of N2 lymph nodes is necessary.Radiomics based on 18F-FDG PET was developed to predict N2 lymph node metastasis in patients with NSCLC.Methods 193 NSCLC patients undergoing ¹⁸F-FDG PET/CT in Tongji hospital of Huazhong University of Science and Technology were retrospectively analyzed.All patients were pathologically confirmed to be lung adenocarcinoma or lung squamous cell carcinoma,and were divided into a training set(n=96)and a validation set(n=97)in a 1:1 ratio.The mediastinal lymph nodes were scored in a 5-point scale using the visual analogue scale(VAS).1132 radiomics features were extracted from PET images using Python platform.The minimum Redundancy Maximum Relevance feature selection algorithm and Least Absolute Shrinkage and Selection Operator were employed to select informative and non-redundant features,and the radiomics signature(Rad-score)was constructed.Differences between groups were tested using Mann-Whitney U test.Multivariate logistic regression was applied to select the important factors.We constructed a combined model based on the clinical variable and radiomics signature,and compared the predictive performance of models using receiver operating curves and Delong test.Results Patients with N2 metastasis had higher Rad-score than those with non-N2 status(training set:z =-4.71,p<0.001;validation set:z =-4.26,p<0.001).The AUCs of Rad-score in the training set and validation set were 0.801(95%CI,0.7070.876)and 0.769(95%CI,0.673-0.849),respectively.By contrast,the AUC values of clinical variables(histology,gender,age,smoking)were relatively lower(training set:0.515,0.508,0.650,0.506;validation set:0.508,0.503,0.531,0.561).Multivariate logistic regression showed that Rad-score(p=0.001),VAS(p<0.001),age(p=0.002)and histology(p=0.035)were independent factors for predicting N2 status.The complex model was constructed according to multivariate logistic regression.AUCs of the composite model for distinguishing N2 status were 0.942(95%CI,0.875-0.979)in the training set and 0.866(95%CI,0.782-0.927)in the validation set.SUVmax,Rad-score,composite model,and VAS were positively correlated with lymph node stations(rho=0.275-0.609,p<0.05),the number of metastatic lymph nodes(rho=0.241-0.603,p<0.05).Rad-score was associated with skipping N2 lymph node metastases(z=-2.074,p=0.038).Conclusion Radiomics from the primary tumor and metabolic information from mediastinal lymph node can effectively predict N2 lymph node metastasis,and combining the radiomics features and visual analogue scale of lymph node score can further improve the diagnostic efficiency.The radiomics based on 18F-FDG PET is expected to be helpful for clinical preoperative staging,treatment strategies and prognosis.Part 4Objective Tumor microenvironment immune types(TMITs)are closely related to the efficacy of immunotherapy.We aimed to assess the predictive ability of ¹⁸F-FDG PET/CT based radiomics of TMITs in pretherapeutic non-small cell lung cancer(NSCLC).Methods A retrospective analysis was performed in 103 patients with NSCLC who underwent ¹⁸F-FDG PET/CT scans in Tongji hospital of Huazhong University of Science and Technology.The patients with NSCLC were divided into a training set and a validation set.Tumor specimens were analyzed by immunohistochemistry for the expression of PD-L1,PD-1,and CD8+TILs;and categorized into four TMITs according to their expression of PD-L1 and CD8+TILs.LIFEx package was used to extract radiomics features.The optimal features were selected using the least absolute shrinkage and selection operator(LASSO)algorithm and a radiomics signature score(Rad-score)was developed.We constructed a combined model based on the clinical variables and radiomics signature,and compared the predictive performance of models using receiver operating curves.Results Four radiomics features(GLRLM LRHGE,GLZLM_SZE,SUVmax,GLCM_Contrast)were selected to build the rad-score.The Rad-score showed a significant ability to discriminate between TMITs in the two sets(p<0.001,0.019),with area under the ROC curve(AUC)equal to 0.800(95%CI,0.688-0.885)in the training set,and 0.794(95%CI,0.615-0.916)in the validation set,compared with 0.738 and 0.699 for the clinical variable.When clinical variables and radiomics signature were combined,the complex model showed better performance in the prediction of TMIT-I tumors,with the AUC increased to 0.838(95%CI,0.731-0.914)in the training set and 0.811(95%CI,0.634-0.927)in the validation set.Conclusion The FDG-PET/CT-based radiomic features showed good performance in predicting TMIT-I tumors in non-small cell lung cancer,providing a promising approach for the choice of immunotherapy in a clinical setting.