| ObjectiveThis study aimed to establish predictive models based on multivariate analyses,and the diagnostic values of models were further assessed by comparing with the quantitative/semi-quantitative parameters in 18F-FDG PET/CT.MethodsWe retrospectively analyzed 94 potentially resectable NSCLC patients who underwent 18F-FDG PET/CT in our Hospital from August 1st,2013 to November 30th,2017.Inclusive criteria:(a)The pathological type was confirmed to be NSCLC;(b)No other malignancies were diagnosed and no cancer-related treatment was accepted before PET/CT;(c)Patients received 18F-FDG PET/CT examination without any tumor-related treatment;(d)N1 and N2 lymph nodes were suspiciously involved(SUVmax>2.5 or short diameter(SAD)≥1cm);(e)No intrapulmonary metastasis,distant metastasis or N3 lymph node involvement were displayed on PET/CT;(f)Complete clinical data were needed.Predictive models were produced using binary logistic regression with a stepwise model-building approach and the Early model was validated.Receiver operating characteristic(ROC)curves were generated to evaluate and compare the diagnostic value of predictive models and quantitative/semi-quantitative parameters(SAD,maximum standardized uptake value(SUVmax),mean standardized uptake value(SUVmean),retention index(RImax)and node-to-background ratio(NBR))and comparison of area under ROC curves(AUCs)was performed by Z scores.The sensitivity and specificity were calculated according to the optimal cut-off values conducted by Youden Index.The difference was statistically significant when P<0.05.ResultsA total of 94 effective cases(67 males and 27 females),with an average of 64±9.2 years(aging from 34 to 84 years)were enrolled in the study,including 165 early nodes and 71 delayed nodes.The area under curve(AUC)and 95%confidence interval(CI)of SAD was 0.805(95%CI:0.738-0.872),and that of SUVmax of early imaging(SUVe-max),SUVmean of early imaging(SUVe-mean)and NBR were 0.807(95%CI:0.738-0.872),0.805(95%CI:0.740-0.873)and 0.793(95%CI:0.772-0.864),respectively.The corresponding optimal cut-off values conducted by Youden Index were of 1cm,6.0,3.0 and 3.5.The binary logistic regression identified the following independent predictors:Age,fuse,SUVmax and SAD.The Early model was established using the following formula:y = ex/(1 + ex),x =3.048 +(2.319 × fuse)-(0.125 × Age)+(0.298 x SUVe-max)+(3.544 × SAD).The AUC of Early predictive model in all nodes was 0.883(95%CI:0.819-0.927),demonstrating better diagnostic value than SAD,SUVe-max,SUVe-mean,NBR(P<0.05).In suspiciously involved nodes,Early predictive model performed better than parameters above,although these differences were not statistically significant.When it came to the lymph nodes with SAD<1cm,both Early model and parameters showed poor diagnostic value.Delayed predictive model was further established as follow:y= ex/(1 + ex),x=5.391-(0.163 x age)+(2.554 x fuse)+(0.327 × SUVd-max)+(4.302 × SAD).The AUCs of Early model,Delayed model,SUVd-max and SUVd-mean were 0.876(95%CI:0.973-0.958),0.888(95%CI:0.806-0.969),0.841(95%CI:0.743-0.939)and 0.801(95%CI:0.693-0.910),respectively,and the differences were not statistically significant.Lymph nodes in the mediastinal 4 and 7 showed negative predictive values(NPVs)of 58.9%and 11.1%when diagnosed with SAD,while lymph nodes in 10-11 with an NPV of 35.3%when diagnosed with SUVe-max.NBR can be used as a good diagnostic indicator for lymph nodes in mediastinal 4,which exhibited up to 100%sensitivity and specificity in this study.Data from December 2017 to February 2018 was verified by Early predictive model.The sensitivity,specificity,positive predictive value(PPV)and negative predictive value(NPV)were 85.7%,100%,100%and 86.7%,respectively.The diagnostic accordance rate was 92.6%.Conclusion(1)Early parameters(SAD,SUVe-max,SUVe-mean,NBR)showed a relatively limited diagnostic value in all intrathoracic lymph nodes,and the corresponding optimal cut-off values were 1cm,6.0,3.0 and 3.5;False positive results were prone to occur in regions 4 and 7 when diagnosed with SAD,and region 10-11 when diagnosed with SUVe-max;NBR was of important reference value in the mediastinal 4 lymph nodes.(2)In early nodes,Early predictive model demonstrated better diagnostic value than early parameters(SUVe-max,SUVe-mean,SAD and NBR);In suspiciously involved nodes,the AUC of Early model was higher than that of early parameters,although the differences were not statistically significant;In nodes with SAD<1cm,both Early model and early parameters demonstrated poor diagnostic value.(3)Early model exhibited similar diagnostic value with Delayed model and.delayed parameters(SUVd-max,SUVd-mean).(4)High diagnostic value of Early predictive model for the differentiation between benign and malignant lymph nodes was further justified by the validation test. |
PDF Full Text Request