| Non-small cell lung cancer(NSCLC)is the common malignant tumor in the clinical practice,accounting for about 85%of lung cancer,and the overall prognosis is poor.Locally advanced non-small cell lung cancer(LA-NSCLC)refers to stage III NSCLC。For nonoperative LA-NSCLC patients treated with chemoradiotherapy,the 5-year survival rate is about 15%.It is a possible way to improve the prognosis that patients are treated with individualized treatment based on the prognosis difference of LA-NSCLC patients treated with chemoradiotherapy.It is the key research direction of precision medicine that predicting the prognosis of LA-NSCLC patients treated with chemoradiotherapy based on radiomics features.Most LA-NSCLC patients treated with chemoradiotherapy have disease recurrence.The recurrence patterns include locoregional recurrence(LR),distant metastasis(DM)and both of LR and DM(LRDM).It is generally believed that LA-NSCLC patients with disease recurrence after chemoradiotherapy can’t be cured.For improving the cure rate,it is necessary to reduce the disease recurrence for LA-NSCLC patients treated with chemoradiotherapy.It is the possible ways to take precision treatment for LA-NSCLC patients treated with chemoradiotherapy before the disease recurrence.It is the important prerequisite for precision treatment to construct the model to predict the recurrence patterns of LANSCLC patients treated with chemoradiotherapy.18Fluorine-Fluorodeoxyglucose positron emission tomography/computed tomography(18F-FDG PET/CT)is one common imaging method in NSCLC clinical practice.18F-FDG PET/CT images can display the spatial and metabolic information of tumor tissues,which is quantified as radiomics features.18F-FDG PET/CT radiomics features include quantitative and semi-quantitative features.Quantitative features are the mathematical expression of pixel or voxel values and their positional relationship.Semi-quantitative features refer to 18F-FDG standardized uptake value(SUV)and its derived features,including metabolic tumor volume(MTV).Many studies showed that 18F-FDG PET/CT radiomics features could predict the prognosis of LA-NSCLC patients treated with chemoradiotherapy.Some studies have shown that 18F-FDG PET/CT radiomics features were significantly associated with the recurrence patterns of LA-NSCLC patients treated with chemoradiotherapy.In this study,multi-centre data was used to construct the models for predicting the prognosis of LA-NSCLC patients treated with chemoradiotherapy by combining 18FFDG PET/CT radiomics features with clinical characteristics,which were compared to seek the best model,and verify the models.Moreover,it was also explored that the correlation between 18F-FDG PET/CT radiomics features,clinical characteristics and the recurrence patterns of LA-NSCLC patients treated with chemoradiotherapy.The model for predicting recurrence patterns was constructed and verified by using 18FFDG PET/CT radiomics features and clinical characteristics.Part Ⅰ Research of 18F-FDG PET metabolic tumor volumes and their dynamic changes to predict prognosis for locally advanced non-small cell lung cancer patients treated with chemoradiotherapyObjectiveTo analyse whether 18F-FDG PET MTVs and their dynamic changes combined with clinical characteristics can predict the prognosis of LA-NSCLC patients treated with chemoradiotherapy.MethodsAccording to the inclusion and exclusion criteria,LA-NSCLC patients treated with chemoradiotherapy in multi-centres were included.All the patients underwent preradiotherapy 18F-FDG PET/CT scanning and chemoradiotherapy,of which some patients underwent mid-radiotherapy 18F-FDG PET/CT scanning.The patients were followed up to obtain progression free survival(PFS)and overall survival(OS).CIGITA was used to delineate the ROIs of primary tumors and both of primary tumors and metastatic lymph nodes displayed by pre-and mid-radiotherapy 18F-FDG PET.Delineate the pre-and mid-radiotherapy primary tumor as ROI1 and ROI2,the pre-and mid-radiotherapy primary tumors and metastatic lymph nodes as RO13 and ROI4.Calculate the MTVs corresponding to ROI1～ROI4:MTV1～MTV4.Calculate the difference(ΔMTV1)of MTV1 minus MTV2 and the difference(△MTV2)of MTV3 minus MTV4.Clinical characteristics,MTV1～MTV4,△MTV1 and MTV2 were used to construct Cox regression equations for predicting PFS and OS.X-tile software was used to determine the cut-off value of independent variable characteristics in multivariate Cox regression equations.Patients with less than or equal to the cut-off value were divided into low value group and the other patients were divided into high value group.K-M survival analysis was used to compare the survival outcomes of patients in low value group and high value group.Results1.One hundred and sixty-five LA-NSCLC patients were included in the analysis.Stage ⅢA,ⅢB and ⅢC patients were 49,85 and 31.The median age,radiotherapy dose and chemotherapy were 62 years old,60.00Gy and 4 cycles.The median PFS and OS were 13.3 months and 32.0 months.2.One hundred and sixty-five patients had pre-radiotherapy 18F-FDG PET/CT images,of which 103 patients had mid-radiotherapy 18F-FDG PET/CT images.The mean±standard deviation of MTV1～MTV4 were 81.92±91.85cm3,36.49 ±48.15cm3,103.55±110.49cm3 and 44.93±52.87cm3.The mean±standard deviation of ΔMTV1 and △MTV2 were 47.48±76.93cm3and 65.59±98.10cm3.3.Univariate Cox regression analysis showed that clinical stage,radiotherapy dose,age,MTV3 and ΔMTV2 were significantly associated with PFS(P<0.05).Clinical stage,radiotherapy dose,chemotherapy cycles,histology,smoking index and MTV1～MTV4 were significantly associated with OS(P<0.05).Multivariate Cox regression analysis showed that radiotherapy dose,age and MTV3 were the independent prognostic characteristics for PFS(P<0.05).Radiotherapy dose and MTV2 were the independent prognostic characteristics for OS(P<0.05).4.Patients were divided into 2 groups with the cut-off value of MTV3=210.60cm3,who with less than or equal to 210.60cm3 were included in the low value group and the other were included in the high value group.PFS in the low value group was significantly better than that in the high value group(P=0.022).Patients were divided into 2 groups with the cut-off value of MTV2=19.50cm3,who with less than or equal to 19.50cm3 were included in the low value group and the other were included in the high value group.OS in the low value group was significantly better than that in the high value group(P=0.000).ConclusionMTVs displayed by 18F-FDG PET images have important prognostic value for LA-NSCLC patients treated with chemoradiotherapy.MTV3 combined with clinical stage,radiotherapy dose and age can predict PFS for LA-NSCLC patients treated with chemoradiotherapy and MTV2 combined with radiotherapy dose can predict OS.Part Ⅱ Research of comprehensive quantitative values of 18F-FDG PET/CT radiomics features and metabolic tumor volumes to predict prognosis for locally advanced non-small cell lung cancer patients treated with chemoradiotherapy ObjectiveTo analyse whether comprehensive quantitative values(CVs)of 18F-FDG PET/CT radiomics features combined with MTVs and clinical characteristics can predict the prognosis of LA-NSCLC patients treated with chemoradiotherapy.MethodsThe patients included in the study were the same as those in the first part of the study and were divided into training set and validation set based on the different centers.The training set and validation set were used to construct and verify the model for predicting prognosis.All the patients underwent pre-radiotherapy 18F-FDG PET/CT scanning and chemoradiotherapy,of which some patients underwent mid-radiotherapy 18F-FDG PET/CT scanning.The patients were followed up to obtain progression free survival(PFS)and overall survival(OS).The model for predicting the prognosis based on the training set was detailed as below.They were the same as those in the first part of the study that region of interest(ROI)1～ROI4 were delineated and MTV1～MTV4,ΔMTV1,ΔMTV2 were calculated.ITK-SNAP software was used to delineate the pre-and mid-radiotherapy primary tumor displayed by 18F-FDG PET as ROI5 and ROI6,the pre-and mid-radiotherapy primary tumor displayed by CT as ROI7 and ROI8.The quantitative values of ROI1,ROI2 and ROI5-ROI8 radiomics features were calculated.Principal component analysis was used to calculate the CVs of ROI1,ROI2 and ROI5-ROI8 radiomics features.The CVs with eigenvalue greater than 1 were included in further analysis.The integrated texture parameter(ITP)values were calculated by using the calculation formula of ITP obtained in our previous research.The clinical characteristics,MTV1～MTV4,ΔMTV1,ΔMTV2,included CVs and ITP were used to construct Cox regression equations for predicting PFS and OS.X-tile software was used to determine the cut-off values of independent variables in the multivariate Cox regression equations.The patients with less than or equal to the cutoff value were divided into low value group and the other were divided into high value group.K-M survival analysis was used to compare the survival outcomes of patients in low value group and high value group.The methods for verifying the prognosis model based on the validation set were detailed as below.Apply the above methods to obtain the CVs and clinical characteristics of patients in the validation set.Substitute the corresponding CVs and clinical characteristics into the multivariable Cox regression equations.Calculate the concordance indexes(C-indexes)when the multivariable Cox regression equation was used to predict PFS and OS in the validation set.Results1.One hundred and sixty-five LA-NSCLC patients included in the analysis were the same as the first part of the study,including 120 patients in training set and 45 patients in validation set.For the training set,the median age,radiotherapy dose,and chemotherapy were 62 years old,60.00Gy,and 4 cycles.The median PFS and OS were 13.4 months and 31.3 months.For the validation set,the median age,radiotherapy dose,and chemotherapy were 61 years old,60.25Gy and 4 cylces.The median PFS and OS were 12.9 months and 40.0 months.2.For the training set,120 patients had pre-radiotherapy 18F-FDG PET/CT images,of which 85 patients had mid-radiotherapy 18F-FDG PET/CT images.The mean ±standard deviation of MTV1～MTV4 were 86.56±98.61cm3,35.93 ± 48.06cm3,108.32±116.08cm3 and 43.55±50.96cm3.The mean ± standard deviation of ΔMTV1 and Δ MTV2 were 46.63±81.29cm3 and 60.23±97.62cm3.Obtain 72 quantitative values from ROI1 and ROI2 respectively,86 quantitative values from ROI5 and ROI6 respectively and 100 quantitative values from ROI7 and ROI8 respectively.Ten CV1(CV1-1～CV1-10),9 CV2(CV2-1～CV2-9),9 CV5(CV5-1～CV5-9),7 CV6(CV6-1～CV6-7),10 CV7(CV7-1～CV7-10)and 11 CV8(CV8-1～CV8-11)were included in further analysis.3.For the training set,univariate Cox regression analysis showed that clinical stage,radiotherapy dose,age,MTV3 and 6 CVs were significantly associated with PFS(P<0.05),and clinical stage,radiotherapy dose,MTV1～MTV4 and 12 CVs were significantly associated with OS(P<0.05).Multivariate Cox regression analysis showed that radiotherapy dose,age and 2 CVs(CV1-9 and CV8-5)were independent prognostic characteristics for PFS(P<0.05).Radiotherapy dose and 4 CVs(CV1-2,CV1-3,CV1-9 and CV8-2)were independent prognostic characteristics for OS(P<0.05).4.The K-M analysis results for the training set patients were shown in the below.PFS and OS were significantly longer for the patients with radiotherapy dose greater than 60Gy than those with radiotherapy dose less than or equal to 60Gy(P<0.05).PFS was significantly longer for the patients with over 60 years old than those with less than or equal to 60 years old(P<0.05);PFS was significantly longer for the patients with CV1-9 greater than 0.31 than those with CV1-9 less than or equal to 0.31(P<0.05);PFS was significantly longer for the patients with CV8-5 greater than-0.75 than those with CV8-5 less than or equal to-0.75(P<0.05).OS was significantly longer for the patients with CV1-2 greater than-0.45 than those with CV1-2 less than-0.45(P<0.05);OS was significantly longer for the patients with CV1-3 less than or equal to 0.20 than those with CV1-3 greater than 0.20(P<0.05);OS was significantly longer for the patients with CV1-9 greater than 0.04 than those with CV1-9 less than or equal to 0.04(P<0.05);OS was significantly longer for the patients with CV8-2 less than or equal to-0.35 than those with CV8-2 greater than-0.35(P<0.05).5.All the C-indexes were 0.542 at 1,2 and 3 years for using multivariate Cox regression equation to predict PFS in the validation set,and all the C-indexes were 0.770 at 1,2 and 3 years for predicting OS.ConclusionThe efficiency of 18F-FDG PET/CT radiomics features CVs is better than that of MTVs for predict the prognosis of LA-NSCLC patients treated with chemoradiotherapy.The efficiency of CV1-9 and CV8-5 combined with radiotherapy dose and age is poor for predicting PFS of LA-NSCLC patients treated with chemoradiotherapy.The efficiency of CV1-2,CV1-3,CV1-9 and CV8-2 combined with radiotherapy dose is preferable for predicting OS of LA-NSCLC patients treated with chemoradiotherapy.Part Ⅲ Research of comprehensive quantitative values of 18F-FDG PET/CT radiomics features and metabolic tumor volumes to predict recurrence patterns for locally advanced non-small cell lung cancer patients treated with chemoradiotherapyObjectiveTo analyse whether comprehensive quantitative values(CVs)of 18F-FDG PET/CT radiomics features combined with MTVs and clinical characteristics can predict the recurrence patterns of LA-NSCLC patients treated with chemoradiotherapy.MethodsLA-NSCLC patients with recurrence patterns in the first part of the study were included in this part and were divided into training set and validation set based on the different centers.The training set and validation set were used to construct and verify the model for predicting recurrence patterns.All the patients underwent preradiotherapy 18F-FDG PET/CT scanning and chemoradiotherapy,of which some patients underwent mid-radiotherapy 18F-FDG PET/CT scanning.The patients were followed up to obtain the recurrence patterns,including LR,DM and LRDM.It was detailed as below that constructing the model for predicting recurrence patterns in the training set.They were the same as those in the first part of the study that region of interest(ROI)1 and ROI3 were delineated and MTV1 and MTV3 were calculated.They were also the same as those in the second part of the study that ROI5 and ROI7 were delineated and the quantitative values of ROI1,ROI5 and ROI7 were calculated.Principal component analysis was used to calculate the CVs(CV9,CV10 and CV11)of ROI1,ROI5 and ROI7 radiomics features respectively.The CVs with eigenvalue greater than 1 were included in further analysis.Spearman correlation analysis or association analysis was used to analyse whether clinical characteristics,MTV1,MTV3,included CVs were significantly associated with recurrence patterns of the patients.Then the characteristics with significant correlation or association coefficients were used to construct the logistic regression equations for predicting the recurrence patterns.The probability values of recurrence patterns were calculated by using the logistic equations.The receiver operating characteristic(ROC)curves for diagnosing recurrence patterns were constructed based on the probability values.The areas under ROC curves(AUCs),diagnostic cut-off values,sensitivities and specificities at the maximum value of Youden index were calculated.It was detailed as below that verifying the model for predicting recurrence patterns in the validation set.Using the above methods to obtain the CVs and clinical characteristics of the validation set patients.Substitute the corresponding CVs and clinical characteristics into the multivariable logistic regression equation.ROC curves were constructed based on predicted recurrence patterns and actual recurrence patterns,then calculate AUCs,sensitivities and specificities.Results1.Eighty-six LA-NSCLC patients were included in the analysis.The training set and validation set were 59 and 27 patients.For the training set,the median age,radiotherapy dose and chemotherapy were 60 years old,60.00Gy and 4 cycles.The median PFS and OS were 9.4 months and 31.0 months.For the validation set,the median age,radiotherapy dose and chemotherapy were 58 years old,60.00Gy and 4 cycles.The median PFS and OS were 9.3 months and 40.4 months.2.For the training set,all the 59 patients had pre-radiotherapy 18F-FDG PET/CT images,of which 32 patients had mid-radiotherapy 18F-FDG PET/CT images.6 of the 32 patients developed LRDM,so that the robust logistic equation for predicting the recurrence events could not be constructed by using the mid-radiotherapy 18F-FDG PET/CT radiomics features.Therefore,the mid-radiotherapy 18F-FDG PET/CT radiomics features were not included in the analysis.For the training set,the mean±standard deviation of MTV1 and MTV3 were 88,97±114.79cm3 and 118.39±136.91cm3.72,86 and 100 quantitative values of 18F-FDG PET/CT radiomics features were obtained from ROI1,ROI5 and ROI7 respectively.3.For the training set,the included CVs in further analysis were 9 CV9(CV9-1～CV9-9)、9 CV10(CV10-1～CV10-9)and 10 CV11(CV11-1～CV11-10).Histology type and 3 CVs(CV9-4,CV11-5 and CV11-7)were significantly associated with recurrence patterns of the patients(P<0.05).Histology type,CV9-4 and CV11-5 as independent variables constituted the logistic regression equations for predicting the recurrence patterns(P<0.05).4.For the training set,the AUC for diagnosing LR was 0.873;taking the probability value greater than 0.560 as the cut-off value,sensitivity and specificity were 0.682 and 0.946.The AUC for diagnosing DM was 0.826;taking the probability value greater than 0.370 as the cut-off value,sensitivity and specificity were 0.750 and 0.829.The AUC for diagnosing LRDM was 0.772;taking the probability value greater than 0.170 as the cut-off value,sensitivity and specificity were 0.846 and 0.609.5.For the validation set,the AUC for diagnosing LR was 0.711;sensitivity and specificity were 0.636 and 0.786.The AUC for diagnosing DM was 0.675;sensitivity and specificity were 0.600 and 0.750.The AUC for diagnosing LRDM was 0.708;sensitivity and specificity were 0.667 and 0.750.ConclusionCVs of 18F-FDG PET/CT radiomics features have an important role in predicting the recurrence patterns of LA-NSCLC patients treated with chemoradiotherapy.CV94,CV11-5 combined with histology type can predict the recurrence patterns of LANS CLC patients treated with chemoradiotherapy. |
PDF Full Text Request