The Clinical Study Of Heterogeneity Of Glucose Metabolism In Non-small Cell Lung Cancer Subtypes And Prognostic Prediction

Part Ⅰ Heterogeneity of glucose metabolism in non-small cell lung cancer subtypesObjective:Adenocarcinoma(ADC)and squamous cell carcinoma(SCC)are the two major histologic subtypes of non-small cell lung cancer(NSCLC).There is a significant difference in the treatment approach,treatment response and prognosis of lung ADC and SCC at different stages.Previous studies have reported that tumor heterogeneity is associated with treatment response and prognosis.However,the effect of clinical staging on the heterogeneity of glucose metabolism of NSCLC subtypes remains unclear.In this study,we investigated the difference in heterogeneity of glucose metabolism among stages Ⅰ-Ⅲ NSCLC subtypes through 18F-FDG PET/CT texture analysis.Methods:Patients who were histologically diagnosed with lung ADC or SCC and underwent pretreatment 18F-FDG PET/CT scans from October 2012 to March 2016 were retrospectively identified.The SUV value of 2.5 was elected as the threshold to automatically delineate the target area of the primary tumor through Chang-Gung Image Texture Analysis(CGITA)software package.The physiological high uptake area was manually eliminated by an experienced thoracic oncologist.In total,6 traditional metabolic parameters and 54 radiomic features describing global(n=5),local(n=27),and regional(n=22)features of the primary tumor were obtained.The differences in radiomic parameters between lung ADC and SCC were compared stage-by-stage in 253 consecutive NSCLC patients with stages Ⅰ to Ⅲ disease.Results:A total of 253 patients were enrolled and analyzed(140 AC and 113 SCC).There was a correlation between metabolic parameters and radiomic features of NSCLC and staging.All six metabolic parameters of the primary tumor were positively correlated with the AJCC stage.Twenty-four radiomic features of 18F-FDG PET/CT imaging of NSCLC were positively correlated with the AJCC stage.Twenty-three radiomic features were negatively correlated with the AJCC stage.Seven radiomic features were not correlated with the AJCC stage.All six metabolic parameters of lung SCC were significantly higher than those of ADC in the same stage.For texture parameters that were positively correlated with the AJCC stage,22,21,and 21 radiomic features of lung SCC were significantly higher than those of ADC in stages Ⅰ,Ⅱ,and Ⅲ,respectively.For texture parameters that were negatively correlated with the AJCC stage,17,14 and 18 radiomic features of SCC were significantly lower than those of ADC in stages Ⅰ,Ⅱ,and Ⅲ,respectively.For the seven radiomic features unrelated to the AJCC stage,only LGLZE in stage I and RP,LGLRE and LGLZE in stage Ⅱ of lung SCC were significantly higher than those of ADC.Conclusion:The difference in glucose metabolic heterogeneity between lung ADC and SCC varied with different stages.It would be more appropriate if prediction studies based on PET imaging were performed in patients with the same pathological types and the same stage.Part Ⅱ Prediction of subtypes of non-small cell lung cancer based on PET radiomicsObjective:It is very crucial to accurately confirm the histological subtype of the NSCLC prior to the treatment decisions.In our previous study,we found that the difference in glucose metabolic heterogeneity between lung ADC and SCC varied with different stages.The aim of this study was to develop a PET radiomic prediction model to distinguish lung ADC from SCC.Methods:Patients who were histologically diagnosed with lung ADC or SCC and underwent pretreatment 18F-FDG PET/CT scans from October 2012 to March 2016 were retrospectively identified as the training cohort.Furthermore,an independent cohort of NSCLC patients who met the inclusion criteria from April 2016 to December 2019 was included as the validation cohort.The SUV value of 2.5 was elected as the threshold to automatically delineate the target area of the primary tumor through CGITA software package.The physiological high uptake area was manually eliminated by an experienced thoracic oncologist.In total,6 traditional metabolic parameters and 54 radiomic features were obtained.The least absolute shrinkage and selection operator(LASSO)algorithm was used for feature selection.A radiomic signature for each stage was subsequently constructed and evaluated.Univariate analysis was performed to select the candidate clinical risk factor.Then,the combined model and nomogram were constructed and evaluated.Receiver operating characteristic(ROC)curve analysis was used to evaluate the predictive effectiveness of the model.And the nomogram was further validated by calibration curve analysis.We assessed the agreement between the predicted and observed tumor subtypes by the Hosmer-Lemeshow test,and a p-value>0.05 indicated good agreement.Results:A total of 416 consecutive patients were analyzed,which included 253 in the training cohort and 163 in the validation cohort.In the training phase of stages Ⅰ,Ⅱ,and Ⅲ,13,5,and 12 out of 60 extracted features were selected with nonzero coefficients.The signatures,also named Rad_Score,of stages Ⅰ,Ⅱ,and Ⅲ are as follows.Rad_Score stage Ⅰ=0.0805*LGSRE+0.2446*StrengthNGTDM+0.0875*ZP+0.0253*HGZE-0.5174*IDMCM-0.0468*SUVmin-0.2532*SUVmax0.7096*SUVmean-0.0930*SUVsd-0.1296*CoarsenessTFC+0.0256*IDMTFCCM+0.2016*CodeSimilarity-0.0489*LNE;Rad_Score stage Ⅱ=0.0410*LGZE+0.0252*ContrastCM-0.8377*SUVpeak + 0.0324*CoarsenessTFC+0.2342*MeanconvergenceTFC;Rad_Score stage Ⅲ=-0.0744*RLV+0.1401*ContrastNGTDM-0.1233*BusynessNGTDM-0.1271*SZE+0.3261*ZP+0.3970*LGSZE-0.2040*LGLZE+0.3625*EnergyCM-0.0022*SUVmin0.8940*SUVmean-0.3982*SUVpeak-0.1066*EnergyTFCCM.The effectiveness of the radiomic signature for predicting lung ADC and SCC in both the training and validation cohorts was good,with AUCs of 0.88,0.85,and 0.90 in the training cohort and 0.93,0.94,and 0.89 in the validation cohort for stages Ⅰ,Ⅱ,and Ⅲ NSCLC,respectively.After univariate analyses,only location served as an independent clinical predictor integrated into the nomogram.The radiomic-clinical nomogram integrating radiomic features with independent clinical predictors showed better predictive performance,with AUCs of 0.98,0.96,and 0.98 in the training cohort and 0.99,0.98,and 0.98 in the validation cohort for stages Ⅰ,Ⅱ,and Ⅲ,respectively.The Hosmer-Lemeshow test further demonstrated the good predictive precision of the nomogram.Conclusion:PET radiomic prediction models provided more favorable performance for discriminating the histological subtype of NSCLC.PET-based radiomic prediction models may help clinicians improve the histopathologic diagnosis of lung cancer in a noninvasive manner.PART Ⅲ Prognostic value of PET radiomic features in patients with locally advanced non-small cell lung cancerObjective:There is a great difference in the treatment response and prognosis of patients with locally advanced NSCLC.If the prognosis of patients can be stratified early and individualized treatment,it is expected to improve the overall survival of patients.The radiomic features derived from neighborhood gray-tone difference matrix(NGTDM)may be associated with differences in response to treatment and survival in NSCLC.The aim of this study was to investigate the prognostic value of PET radiomic features derived from NGTDM in patients with locally advanced NSCLC who were treated with concurrent chemoradiotherapy.Methods:The patients with locally advanced NSCLC from January 2014 to December 2018 were retrospectively identified.These patients were treated with concurrent chemoradiotherapy and underwent pretreatment 18F-FDG PET/CT scans.Metabolic parameters and radiomic features derived from NGTDM were extracted through the CGITA software package.Overall survival(OS)and progression-free survival(PFS)were recorded.Cox regression analyses were performed to identify prognostic factors for PFS and OS.ROC curves analysis was used to determine the optimum cutoff value of metabolic parameters and radiomic features for survival.Results:62 patients with locally advanced NSCLC who were treated with concurrent chemoradiotherapy were included in the study.MTV,contrast,busyness,complexity,and strength were significantly different between responders and nonresponders.Median OS was 27.60 months(95%CI,22.85-30.60 months),PFS was 14.75 months(95%CI,12.35-18.40 months).Cox multivariate regression analysis showed that strength was an independent predictor for PFS,while strength and clinical stage were independent predictors for OS.Kaplan-Meier survival analysis showed there was significant association of PFS with strength.Patients with tumor strength>26.50 had better PFS than those with tumor strength ≤26.50(mean PFS:22.72 vs.15.57 months,P<0.001).Kaplan-Meier survival analysis showed there was significant association of OS with strength and clinical stage.Patients with tumor strength>13.96 had better OS than those with tumor strength≤13.96(mean OS:32.58 vs.23.93 months,P<0.001).Patients with stage ⅢA had better OS than those with stage ⅢB(mean OS:36.22 vs.20.05 months,P<0.001).Conclusion:PET radiomic features derived from NGTDM are associated with treatment response and survival of patients with locally advanced NSCLC who were treated with concurrent chemoradiotherapy.The radiomic features of PET images which can predict the curative effect and prognosis are helpful to the stratification and individualized treatment of NSCLC patients.