| Part ? The comparative study of PET/CT,CT and PET radiomics models in predicting lymph node metastasis of lung adenocarcinoma(diameter?3cm)Objective: The value of 18F-FDG PET/CT,CT and PET radiomics models in the prediction of local lymph node metastasis in lung invasive adenocarcinoma(diameter?3cm).Methods:528 patients with lung adenocarcinoma confirmed by postoperative pathology were analyzed retrospectively.Among them,149 patients had lymph node metastasis,while 379 patients had no lymph node metastasis.All cases were collected under the same equipment and parameters.After image preprocessing,they were randomly divided into training group and test group according to the proportion of 7:3.Intra-and inter-class correlation coefficient(ICC)was used to evaluate sketch consistency.ICC>0.75 indicated that the consistency was good.Using m RMR and LASSO to determine the best feature set,and using logistic regression to establish PET/CT,CT,and PET radiomics models for predicting lymph node metastasis in lung adenocarcinoma.Receiver operating characteristic(ROC)curves evaluates the diagnostic efficacy of three radiomics models to predict the lymph node metastasis of lung adenocarcinoma.The critical value is taken when the Youden index is the largest,and the sensitivity,specificity and accuracy of predicting the lymph node metastasis of lung adenocarcinoma are calculated,and the models were verified in the test group.Results: PET/CT,CT and PET radiomic models' datum were retained 13 PET/CT(5 CT and 8 PET),10 CT,and 10 PET radiomics features after feature extraction and dimensionality reduction,respectively.The AUC values of PET/CT,CT and PET radiomic models in training groups were 0.92,0.87 and 0.83;The AUC values of the test group were 0.92,0.82 and 0.82.The AUC values of PET/CT,CT and PET radiomic models were respectively measured by De Long test,which showed that the AUC value of PET/CT radiomics model was higher than that of CT and PET radiomics model,and the difference between PET/CT and CT radiomic model(P<0.05),PET/CT and PET radiomic model(P<0.05)were statistically significant.The AUC value of CT radiomic model was slightly higher than that of PET radiomic model,but there was no statistical significance between the two models(P value was 0.293).The sensitivity,specificity,and accuracy of PET/CT radiomic model for predicting lymph node metastasis in lung adenocarcinoma training group were 0.805,0.886,and 0.827,respectively;the test group was 0.858,0.932,and 0.879 respectively.The sensitivity,specificity and accuracy of CT radiomic model for predicting lymph node metastasis in the training group were 0.829,0.759,and 0.809,respectively;test groups were 0.718,0.9,and 0.764,respectively.The sensitivity,specificity,and accuracy of PET radiomic model prediction for lymph node metastasis training group were 0.767,0.743,and 0.760,respectively;the test group was 0.770,0.864,and 0.796,respectively.Conclusion: PET/CT radiomics model has good predictive ability in predicting lymph node metastasis of lung adenocarcinoma(diameter?3cm),and it is superior to CT and PET radiomics models.Part ? The diagnostic value of PET/CT radiomics-clinical composite model for predicting lymph node metastasis in lung adenocarcinoma(diameter?3cm)Objective: To explore the value of the PET/CT radiomics-clinical composite model in predicting lymph node metastasis of lung adenocarcinoma(diameter?3cm),and compare the differences of diagnostic efficacy between composite model,radiomics model and the clinically factor model.Methods: The clinical data of 528 cases(same as in part 1)of lung invasive adenocarcinoma were retrospectively analyzed,and a clinical factor model for predicting lymph node metastasis of lung adenocarcinoma(diameter?3cm)was established.In order to verify the reliability and repeatability of the PET/CT radiomics model(same as in part 1),an internal cross-validation was performed 100 times in the training group and the test group.Through logistic regression analysis,the statistical clinical factors which were finally included in the clinical model after screening were combined with PET/CT radiomics model according to the average weight to get PET/CTradiomics-clinically combined model.The above three models were all built in the training group and tested in the corresponding test group.Construct a nomogram of the PET/CT radiomics-clinical model to assist physicians in making clinical decisions.The clinical decision curves of the clinical factor model and the PET/CT radiomics-clinical model were drawn,and the clinical net benefit under different threshold probability was quantified and analyzed in the test group to determine the clinical practicability and application value of the two models.Results: This part of the study showed that two clinical factors,solid component size and location,could be used as independent predictors of lung adenocarcinoma lymph node metastasis.A clinical factor model was obtained through logistic regression analysis,and the AUC value in the training group was 0.91,and the AUC value in the test group was 0.95.The AUC value of the raidomic model in the training group was 0.92,and the test group was 0.92.The AUC values of the composite model in the training set and test set were 0.95 and 0.95,respectively.The AUC value of the composite model was higher than the clinical model,and there was a statistical difference(P <0.05);the AUC value of the composite model was slightly higher than that of the radiomics model,and there was no statistical difference between them(P value was 0.092);the AUC value of radiomics model was slightly higher than the clinical factor model,and there was no statistical difference between the two models(P value was 0.874).The accuracy,sensitivity,and specificity of the PET/CT composite model training group were 0.841,0.649,and 0.977,respectively;the test group was 0.866,0.683,and 0.989.Solid composition size,position and rad-score were used to construct a nomogram of PET/CT radiomics-clinical model.The goodness of fit test curve in the training group and the test group showed that the P values were 0.597 and 0.133,respectively.The clinical decision curve shows that when the threshold probability of predicting lymph node metastasis of lung adenocarcinoma is 1-70%,the use of PET/CT radiomics-clinical model to predict the lymph node metastasis of lung adenocarcinoma has more benefits than the clinical model.Conclusion: This part of the study shows that the radiomics features of PET/CT images based on lung adenocarcinoma(diameter?3cm)can increase the clinical diagnosis efficacy of predicting lymph node metastasis of lung adenocarcinoma.The nomogram of PET/CT radiomics scores combined with clinical features is a clinician's prediction of lung adenocarcinoma lymph node metastasis provides a quantitative and intuitive method.Part ? The comparative study of PET/CT,CT and PET radiomics models in predicting EGFR genotype of lung adenocarcinomaObjective: The value of 18F-FDG PET/CT,CT and PET radiomics models in predicting EGFR gene mutation in lung adenocarcinoma.Methods: A retrospective analysis of 583 patients with postoperative pathologically confirmed lung adenocarcinoma,including 295 patients with lung adenocarcinoma EGFR mutation and 288 patients with wild EGFR.All the patients were examined under the same PET/CT and the same scanning parameters.After image preprocessing,the patients were randomly divided into training set and test set according to the proportion of 7:3.Using m RMR and LASSO for feature extraction,selecting the best feature set,and using logistic regression to establish PET/CT,CT and PET radiomics models for predicting EGFR mutations in lung adenocarcinoma.ROC curves were used to evaluate the diagnostic efficacy of three radiomics models to identify EGFR mutation in lung adenocarcinoma,and the models were verified in the test group.Results:PET/CT,CT and PET radiomics models' datum were retained 18 PET/CT(10 CT and 8 PET),15 CT,and 12 PET radiomics features after feature extraction and dimensionality reduction,respectively.The AUC values of the PET/CT,CT and PET radiomics models in training groups were 0.76,0.74 and 0.71;The AUC values of the test groups were 0.75,0.70 and 0.75.De Long tests were performed between PET/CT,CT and PET radiomic models,showing that the AUC value of PET/CT radiomics model was higher than CT radiomics model,but the two group had no statistical significance(P value 0.098).The AUC value of PET/CT radiomics model was higher than PET radiomics model,and both were statistically significant(P<0.05);the AUC value of CT radiomics model was higher than the PET radiomics model,and there was no statistically significant(P value was 0.113).The sensitivity,specificity and accuracy of PET/CT radiomics model for lung adenocarcinoma EGFR mutation training group were 0.703,0.686,and 0.694,respectively;the test group was 0.535,0.864,and 0.701,respectively.The sensitivity,specificity,and accuracy of CT radiomics model prediction for EGFR mutation training group were 0.609,0.754,and 0.682 respectively;the test group was 0.802,0.568,and 0.684.The sensitivity,specificity,and accuracy of the PET radiomics model predictionEGFR mutation training group were 0.569,0.758,and 0.665;the test group was 0.663,0.773,and 0.718,respectively.Conclusion: PET/CT radiomics model is better than PET radiomics in predicting EGFR mutation of lung adenocarcinoma,which is of certain clinical value in predicting EGFR mutation of lung adenocarcinoma and selecting treatment options.Part ? Study on the diagnostic value of PET/CT radiomics-clinical composed model for predicting EGFR gene phenotype in lung adenocarcinomaObjective: To study on the value of the PET/CT radiomics-clinical composed model in the predicting EGFR mutations of lung adenocarcinoma and compare the differences of diagnostic efficacy between PET/CT radiomics-clinical model,PET/CT radiomics model and the clinical model.Materials and methods: A retrospective analysis of 583 cases of lung adenocarcinoma(same as the cases in part 3)was used to establish a clinical factor model for predicting EGFR mutations in lung adenocarcinoma.In order to verify the reliability and repeatability of the PET/CT radiomics model(same as in part 3),an internal cross-validation was performed 100 times in the training group and the test group.The logistic regression analysis was used to combine the statistically significant clinically factors that were finally included in the clinical model after screening with the PET/CT radiomics model by the average weight,and the PET/CT radiomics-clinically combined model was obtained.The above three models are all built in the training group and tested in the test group.The ROC curves were used to evaluate the PET/CT radiomics-clinical model,clinical model and radiomics model to predict the diagnostic efficacy of lung adenocarcinoma EGFR mutation.In order to assist doctors in diagnosis,a composed model nomogram was constructed.The clinical decision curves of the clinical factor model and PET/CT radiomics-clinical composed model were drawn,and the clinical net benefits under different threshold probabilities were analyzed quantitatively in the test group to determine the clinical utility of the two models in predicting EGFR gene mutation in lung adenocarcinoma.Results: The study showed that 4 clinical factors of smoking history,pleural traction,air bronchi sign and ground glass component can be used as independent predictors of lung cancer adenocarcinoma EGFR mutations.Then a clinical factor model for predicting EGFR mutation in lung adenocarcinoma was obtained through logistic regression analysis.The AUC value of the training group was 0.81,and the AUC value of the test group was 0.77.The AUC values of the combined model in the training and test groups were 0.84 and 0.81,respectively.The AUC value of radiomics model in the training group is 0.76,and the AUC value of the test group is 0.75.The AUC value of the PET/CT radiomics-clinical model is higher than the clinical factor and radiomics model,and there are statistical differences(P<0.05);the AUC value of the clinical factor model is slightly higher than the radiomics model,but there is no statistical difference between the two groups(P value 0.167).The accuracy,sensitivity,and specificity of the PET/CT composed model training group were 0.770,0.751,and 0.793,respectively;the AUC value,accuracy,sensitivity,and specificity of the test group were 0.805,0.875,and 0.755,respectively.The 4 clinical factors and radiomics scores were used to construct a PET/CT radiomics-clinical model alignment chart.The Hosmer-Lemeshow test of the training group and the test group showed that the P values were 0.34 and 0.13,respectively.The DCA curve shows that when the threshold probability of predicting EGFR mutations in lung adenocarcinoma is 10-85%,the use of PET/CT radiomics-clinical model has more benefit in predicting EGFR mutations in lung adenocarcinoma than in the clinical factor model,and it is between 20-50% is better.Conclusion:This part of the study shows that the radiomics features of PET/CT images based on lung adenocarcinoma can increase the clinical diagnosis efficacy of EGFR mutations in lung adenocarcinoma.The nomogram of PET/CT radiomics scores combined with clinical features is a clinician's prediction of lung adenocarcinoma EGFR mutations provides a quantitative and intuitive method. |
PDF Full Text Request