| Objective:To explore the metabolic features,radiomics,and body composition using18F-fluorodeoxyglucose Positron emission tomography computer tomography predict the prognostic value in patients with non-small cell lung cancer treated with chemotherapy.Methods: Patient clinical information and PET/CT images were collected and followed up.The endpoints were progression-free survival(PFS)and overall survival(OS).The effect factors on PFS and OS were analyzed by the Kaplan-Meier method;the survival differences between groups were analyzed by the log rank test;and the COX risk proportional regression model performed a multivariate prognosis analysis.PET and CT the volume of primary area of interest(VOI)delineated by LIFIx and ITK-SNAP,respectively.PET/CT radiomics characterization extraction based on progression / death risk for outcome,Five machine learning classifiers,namely random forest(RF),support vector machine(SVM),logistic regression(LR),e Xtreme Gradient Boosting(XGBOOST)and Natural Gradient Boosting(NGB),were used to construct the radiomics models,To assess the sensitivity of the five models,specificity,precision,Accuracy and the Kappa coefficient,And the efficacy of each model was evaluated by the AUC.The Delong-test was used to evaluate the prediction efficacy of the optimal radiomics model,the clinical model and the combined model.The fat area and skeletal muscle tissue content were segmented by CT images at the third lumbar vertebra(L3)level by slice Omatic software.By height or weight,the body mass index(BMI),subcutaneous and visceral fat volume index(SFVI,VFVI),and skeletal muscle index(SMI)were calculated.A Pearson correlation analysis was used to explore the correlation between body composition and PFS,OS,and the Kaplan-Meier method.The survival differences between groups were analyzed by the log rank test.Results: A total of 146 patients with moderately advanced NSCLC were included in the first and second stages of the study,with PFS of 7.02 months(range: 0.80-55.67months)and OS of 21.80 months(range: 1.70-63.77 months).In the third stage,112 patients were included,with PFS 6.37 months(range 0.80-55.67 months)and OS22.08 months(range 1.70-63.77 months).The large AUC when all PET/CT parameters used PFS as the survival endpoint.The results of univariate analysis by the Kaplan-Meier method showed that pathological type,bone metastasis,MTV30,MTV40,and MTVwb were significantly associated with PFS,and stage,CT signs such as lobsign,burr sign,lesion density uniformity,obstructive pneumonia/obstructive atelectasis,lymph node metastatic,lesion maximum diameter,and PET signs such as SUVmean and MTVwb were significantly associated with OS.The results of COX analysis showed pathological and bone metastasis of PFS and focal obstructive pneumonia,maximum lesion diameter,and SUVmean of OS in advanced NSCLC patients.The model of the five radiomics classifiers based on the risk of progression as the outcome had a sensitivity of 57.75% – 88.73%,the specificity was73.33% – 96%,and the Kappa coefficient was 0.3173 – 0.8217.The AUC was 0.651– 1.0 in the training group and 0.525 – 0.644 in the validation group.In the validation group,the SVM was the best,and the AUC based on the radiomics model,clinical model and composite model were 0.753,0.746 and 0.716,respectively,the Delong test validated the radiomics model and the composite model(Z = 1.777,P = 0.0756 > 0.05)and the clinical model(Z = 0.323,P = 0.746 > 0.05)were not statistically different in their efficacy for predicting progression risk.The classifier sensitivity of the radiomics model based on death risk as outcome was 78.08%-89.04%,with41.10%-93.15% specificity,Kappa coefficient was 0.1398-0.6657,AUC 0.578-1.0 in the training group and 0.409-0.676 in the validation group.The Pearson correlation analysis results showed that the SFVI was positively associated with the median progression survival time(P<0.05).The results of Kaplan-Meier analysis showed that PFS and OS were extended by 5.1 months compared with the sarcopenia group,which was statistically significant.Conclusion:IPathology and bone metastasis were independent prognostic factors for PFS(<0.05).The presence of obstructive pneumonia or nontension,a focal maximum diameter greater than 51.50 mm,and a SUVmean greater than 13.78 were all independent prognostic factors for OS(<0.05).Radiomics models of different machine learning algorithms were different in predicting the prognosis performance of NSCLC patients,and the best radiomics model performed when progression risk was the outcome,in which SVM efficacy was the best,and there was no significant difference in predicting progression risk based on SVM radiomics model,clinical model and combination model.SFVI was positively associated with PFS in the patients,and sarcopenia is a prognostic factor for both PFS and OS in the patients. |
PDF Full Text Request