Study Of ¹⁸F-FDG PET/CT And Radiomics In Predicting Lymph Node Metastasis And Disease-free Survival Of Non-small Cell Lung Cancer

Purpose:1.We investigated whether the ¹⁸F-FDG PET/CT-based radiomics models（RM）could predict pathological mediastinal lymph node metastasis in patients with NSCLC undergoing surgery.2.Tumor promote disease progression by reprogramming their metabolism and that of distal organs,so it is of great clinical significance to study the changes in glucose metabolism at different tumor stages and their effect on glucose metabolism in other organs.3.The purpose of this study was to evaluate the relationship between preoperative tumor SUVmax and skeletal muscle and visceral fat in patients with NSCLC and its effect on DFS.Material and Methods:1.Part A of this study included 716 patients with a clinicopathological diagnosis of NSCLC were included in this retrospective study.The prediction model A was developed in the training cohort A that consisted of 501 patients.Radiomics features were extracted from ¹⁸F-FDG PET/CT of the primary tumor.Support vector machine and extremely randomized trees were used to build the RM-A.Internal validation was assessed.The independent testing cohort A contained the remaining 215 patients.The performances of the RM-A and clinical node staging（c N staging）in predicting node metastasis（p N0 vs.p N1 and N2）were compared for each cohort.The area under the curve（AUC）of the receiver operating characteristic curve（ROC）analysis was applied to assess the performance of RM-A.In part B,patients with obvious enlarge and abnormal ¹⁸F-FDG uptake of mediastinal lymph nodes were excluded,and a total of 570 patients with T1～2 N0～2 NSCLC were included.Among them,the image data of 399 patients constitute the training cohort B,and the data of 171 patients constitute the testing cohort B.2121radiomics features,9 PET parameters and 9 clinicopathological data were collected,a total of 2139 features.T-test or rank sum test,single factor logistic regression analysis,multivariable logistic regression analysis,the gradient boosting algorithm were used to screen the parameters.Finally,there are 7 remaining parameters.The RM-B was trained by logistic regression classification model.The Rad-score of each primary tumor was calculated by the RM-B,and the best cut-off value of the Rad-score for predicting lymph node metastasis was obtained by ROC analysis;Rad-score was integrated with clinicopathological parameters,PET metabolic parameters and c N stage.The best parameter combination was selected as Rad-score and c N stage by multivariable logistic regression analysis,and the nomogram of radiomics combined with clinical parameters was drawn.2.A retrospective single-center study was conducted on 253 NSCLC（non-small cell lung cancer）patients with negative lymph nodes and no distant metastasis.According to the AJCC system,the patients were divided into different groups based on tumor size:group 1,T1stage≤3cm（n=121）;group 2,3cm﹤T2a≤4cm（n=64）;group 3,4cm﹤T2b stage≤5cm（n=36）;and group 4,5cm≤T3 stage﹤7cm（n=32）.All the patients underwent baseline ¹⁸F-FDG PET/CT scans,and the primary lesion SUVmax,liver SUVmean,spleen SUVmean,TLR（Tumor-to-liver SUV ratio）and TSR（Tumor-to-spleen SUV ratio）were included in the study,combined with clinical examination indicators to evaluate DFS.3.154 patients with NSCLC without distant metastasis were studied retrospectively.All patients underwent baseline ¹⁸F-FDG PET/CT scanning,the height and weight data of all patients were collected,and the corresponding body mass index（BMI）was calculated;SUVmax,liver SUVmean and spleen SUVmean of primary tumors were measured by aw workstation.The skeletal muscle area（SMA,cm²）,skeletal muscle mass index（SMI）,skeletal muscle radiation density（SMD,HU）,visceral fat area（VFA,cm²）,visceral adipose tissue index（VATI）and skeletal muscle visceral fat ratio（SVR）were measured by Image J software.Kaplan-Meier survival analysis was performed for DFS in combination with the above parameter evaluation.Results:1.The AUC of the RM-A（0.81,95%CI:0.771,0.848;sensitivity,0.794;specificity,0.704）for the predictive performance of lymph node metastasis was significantly better than that of c N in the training cohort A（0.685,95%CI:0.644,0.728;sensitivity,0.804;specificity,0.568）,（P value=8.29e-07,as assessed by Delong test）.In the testing cohort B,the AUC of the RM-A（0.766,95%CI:0.702-0.830;sensitivity,0.688;specificity,0.704）was also significantly higher than that of c N（0.685,95%CI:0.619,0.747;sensitivity,0.799;specificity,0.568）,（P=0.0371,Delong test）.The RM-B was developed in the training cohort B that consisted of 399 patients（122 patients with N1/2 stage,277 N0 stage）.52patients with pathological N1/2 stage and 119 patients with pathological N0 stage were included in the testing cohort B.Through ROC analysis,the best cut-off value of Rad-score calculated by radiomics model is-0.85.The accuracy,AUC,sensitivity and specificity of Rad-score in training cohort B and testing cohort B are 0.682,0.754（95%CI:0.713,0.793）,0.746 and 0.653;0.655,0.72（95%CI:0.649,0.79）,0.635 and 0.664respectively.Then the training cohort B and testing cohort B were combined to compare the classification performance of Rad-score and c N stage.The accuracy,AUC,sensitivity and specificity were 0.68,0.74（95%CI:0.71,0.78）,0.71 and 0.66;0.71,0.63（95%CI:0.60,0.67）,0.44and 0.82 respectively.Integrating Rad-score with clinicopathological parameters and c N stage,the best parameter combination was selected as Rad-score and c N stage by multivariable logistic regression analysis,and the radiomics clinical nomogram was obtained.The accuracy,AUC,sensitivity,and specificity were 0.726,0.756（95%CI:0.719,0.792）,0.647 and 0.76 respectively.Finally,the Delong test was used to compare the efficacy of radiomics clinical nomogram,Rad-score and c N stage.It showed that the p-value of the three comparisons were less than 0.05.It can be seen that the radiomics clinical nomogram combined with Rad-score and c N stage further improved the ability to predict mediastinal lymph node metastasis of NSCLC.2.In NSCLC patients,with the increase in the maximum diameter of the tumor,the SUVmax of the primary lesion gradually increased,and the SUVmean of the liver gradually decreased.The primary lesion SUVmax,liver SUVmean,TLR and TSR were related to disease recurrence or death.The best predictive parameters were different when the tumor size differed.SUVmax had the highest efficiency when in T1 and T2a stage（T1,AUC:0.707 95%CI,0.430-0.984;T2a,AUC:0.72,95%CI,0.539-0.912）.Liver SUVmean had the highest efficiency in T2b stage（AUC:0.712,95%CI,0.535-0.889）,and TLR had the highest efficiency in T3 stage（AUC:0.925,95%CI,0.820-1.000）.3.Compared with the low SUVmax group of primary tumors,the mean values of BMI,SMA,VFA and VATI in the high SUVmax group were significantly higher.There were also significant differences in histopathological type,degree of pathological differentiation,AJCC stage and T stage between the two groups.Patients with sarcopenia were older and had lower BMI,SMA and SMI.The proportion of men in H-VFA group was higher,and their BMI,SMA,VFA,VATI,SVR and liver SUVmean were higher.Univariate analysis of DFS showed that VFA,VATI,degree of pathological differentiation,tumor SUVmax,AJCC stage,T stage and N stage all affected DFS,while the parameters reflecting skeletal muscle content had nothing to do with DFS.Multivariate regression analysis found that only VFA and SUVmax were related to DFS.Kaplan-Meier survival analysis showed that high SUVmax,low VFA,higher T stage and higher N stage were related to the reduction of DFS.Comparing the survival curve drawn by the combination of tumor SUVmax and VFA with AJCC staging,it was found that tumor SUVmax combined with VFA could better distinguish the DFS of patients,and the DFS of patients with low tumor SUVmax and H-VFA was significantly better.Conclusions:1.The RM-A based on ¹⁸F-FDG PET/CT has potential for p N staging in patient with NSCLC.The clinical radiomics nomogram can significantly benefit patients in predicting mediastinal lymph node metastasis of NSCLC.2.In patients with early NSCLC,glucose metabolism reprogramming occurs in the primary lesion and liver.With the increase in tumor size,different metabolic parameters should be selected to evaluate the prognosis of patients.3.Preoperative ¹⁸F-FDG PET/CT could comprehensively evaluate the SUVmax,skeletal muscle and visceral fat of patients with NSCLC.Compared with traditional AJCC staging,the combination of primary tumor SUVmax and VFA can better distinguish postoperative DFS in patients with early NSCLC.