| Part Ⅰ Correlation between Metabolic Parameters of Primary Lesion in 18F-FDG PET-CT and Lymph Node Metastasis,Visceral Pleural Invasion and Lymphovascular Invasion in Peripheral Clinical T1 Non-small-cell Lung CancerObjective:Studies have shown that sublobar resection(segmentectomy,or wedge resection)is suitable for patients with peripheral clinical T1N0M0 non-small-cell lung cancer(NSCLC).However,sublobar resection is not suitable for NSCLC with pathological invasiveness in peripheral clinical T1(cT1)stage(including lymph node metastasis(LNM),visceral pleural invasion(VPI),lymphovascular invasion(LVI)).Therefore,preoperative evaluation or screening NSCLC patients with pathological invasiveness in peripheral cT1 is of paramount importance for the choice of surgical methods.Previous studies have confirmed that the maximum standardized uptake value(SUVmax)of NSCLC primary lesions on 18F-FDG PET-CT is related to pathological invasiveness,but the difference of SUVmax measurement accuracy between different equipment and different scanning schemes is still an urgent problem to be solved.The purpose of this study was to investigate the dose-effect relationship between the metabolic parameters of the primary lesions and the pathological invasiveness of primary lesions in 18F-FDG PET-CT of patients with peripheral cT1 stage NSCLC.Methods:This study was a retrospective analysis.The patients underwent 18F-FDG PET-CT examination before surgery and were clinically diagnosed as clinical T1 lung cancer.All patients received radical resection of the lesions with lymph node dissection in our center within one month after examination,and the postoperative pathological results were taken as the "gold standard".The pathological results confirmed that the primary lesion had any of the following characteristics(such as LNM,VP I or LVI)was defined as invasiveness group,and those without any of the above characteristics were defined as less-invasiveness group.The differences of clinical variables,pathological variables,metabolic parameters and morphological parameters of the primary lesion between invasiveness group and less-invasiveness group were compared.The exposure factor in this study was set as the primary lesion standard uptake rate(SUR)standardized by the mediastinal blood pool.The SUR of primary lesions standardized by descending aortic mediastinal blood pool was divided into three equal parts according to the sample size and sample distribution,and the above parameters were tested according to the trinket of SUR.The confounding factors were determined according to the results of statistical differences and trend tests between the invasiveness group and less-invasiveness groups.And then univariable and multivariable logistic regression models(including unadjusted,simply adjusted general information model and fully adjusted confounding factor model)were used to calculate odds ratio(OR)and 95%confidence interval(95%CI)to evaluate the dose-effect relationship between SUR and pathological invasiveness of primary lesions after considering all confounding factors.Finally,the generalized additive model(GAM)was used to fit the smooth curve between SUR and pathological invasiveness.Results:From January 1,2017 to March 31,2020,a total of 2412 patients who underwent preoperative 18F-FDG PET-CT examination in the Department of Nuclear Medicine of the third affiliated Hospital of Soochow University and were clinically diagnosed as clinical T1 lung cancers were analyzed retrospectively.According to the inclusion and exclusion criteria,174 patients met the inclusion criteria finally..A total of 67 patients(38.5%)had pathological invasiveness,and the results showed that the SUR value of primary lesions in the invasiveness group was higher than that in the less-invasiveness group[6.50(4.82-11.16)vs.4.12(2.04-6.61),P<0.001].The trend tests of SUVmax,mean standardized uptake value(SUVmean),metabolic tumor volume(MTV),total lesion glycolysis(TLG),mean CT value(CTmean),primary lesion size,neuron-specific enolase(NSE),LNM proportion,adenocarcinoma proportion,poor differentiation and SUR value were statistically significant(P<0.001,<0.001,0.010,<0.001,<0.001,0.002,0.033,<0.001,0.002 and<0.001,respectively).Univariate logistic regression analysis showed that the risk of pathological invasiveness increased significantly with the increase of primary lesion’s SUR[OR:1.13(95%CI:1.06-1.21),P<0.001].The results of multivariable logistic regression model analysis showed that when the primary lesion’s SUR was a continuous variable,it was still significantly correlated with pathological invasiveness after considering all confounding factors[OR:1.09(95%CI:1.01-1.18),P=0.032].GAM fitting curve showed that there was a positive linear correlation between SUR and pathological invasiveness of primary lesion.Finally,the E-value analysis results(E value=1.26)showed that the results were robust to the all unmeasured potential confounding factors.Conclusions:In patients with peripheral cT1 stage NSCLC,after considering all confounding factors,there is a linear positive correlation between the SUR value of the primary lesion and the pathological invasiveness of the primary lesion,which can be used as a supplementary risk factor to assess the risk of pathological invasiveness of the primary lesion of peripheral cT1 stage NSCLC.Part Ⅱ Clinical Study in Metabolic Parameters of Primary Lesion in 18F-FDG PET-CT and CT Morphological Parameters in Predicting Lymph Node Metastasis of Peripheral Clinical T1 Non-small-cell Lung CancerObjective:According to the conclusion from several large randomized controlled clinical trials of the Japanese Clinical Oncology Group,the sublobar resection has the advantages of less complications and good prognosis in patients with cTIN0M0 stage NSCLC.Therefore,the use of preoperative imaging methods to accurately find the patients with cTIN0M0 stage NSCLC is still essential for clinicians to choose the mode of operation.As a preoperative clinical staging method recommended by the American College of Chest Physicians(ACCP)guidelines,18F-FDG PET-CT has the greatest advantage of combining anatomical information with metabolic information.However,the traditional 18F-FDG PET-CT diagnostic criterion based on hilar or mediastinal lymph node short diameter≥1.0 cm or lymph node SUVmax≥ 2.5 have low diagnostic performance,and studies have shown that its sensitivity and negative predictive value(NPV)were 47.4%and 87.7%,respectively.Therefore,the primary aim of this study was to explore the clinical role of preoperative metabolic parameters and morphological parameters of primary lesion in predicting LNM in peripheral clinical stage T1 NSCLC.Methods:From January 1,2017 to March 31,2020,patients who underwent preoperative 18F-FDG PET-CT examination in the Department of Nuclear Medicine of the third affiliated Hospital of Soochow University and were clinically suspected as clinical T1 lung cancers were analyzed retrospectively.According to postoperative pathology,the patients were divided into two groups of the LNM group and the non-LNM group.The gender,age,height,weight,smoking history(negative or positive),the number of lymph node dissection and surgical procedure(lobectomy or sublobar resection)were collected.The maximum standard uptake value(SUVmax),mean standard uptake value(SUVmean),tumor metabolic volume(MTV),total lesion glycolysis(TLG)of the primary lesion on PET images,and the CTmean of primary lesions on CT images were measured.And edge information(lobulation sign,spiculation,pleural depression)of the primary lesions on CT images were analyzed respectively.The differences of clinical variables,metabolic parameters and morphological parameters of primary lesions between the two groups were compared.Then the independent risk factors of LNM of peripheral clinical stage T1 NSCLC were determined by univariate analysis and multivariate analysis.The receiver operating characteristic curve(ROC)combined with independent risk factors and traditional 18F-FDG PET-CT diagnostic criterion were drawn respectively,and the area under the curve(AUC)was compared by De-long test to evaluate the diagnostic efficacy.Then the sensitivity,specificity,positive predictive value(PPV)and NPV were calculated respectively.Finally,the ten-fold cross-validation method and the goodness-of-fit test were used to verify the model established by the combined independent risk factors.Results:From January 1,2017 to March 31,2020,a total of 2412 patients diagnosed as clinical T1 lung cancers were examined by 18F-FDG PET-CT in the Department of Nuclear Medicine of the third affiliated Hospital of Soochow University.According to postoperative pathology,the incidence of LNM was 24.7%.In the LNM group,the SUVmax,SUVmean,CTmean,the maximum diameter of the lesion and TLG of the primary lesions were significantly higher than those in the non-LNM group.Univariate analysis showed that SUVmax,SUVmean,mean CT value,size and TLG of primary lesion preoperation were significantly correlated with LNM,ORs(95%CIs)and p-values were(1.09(1.04-1.15);p<0.001),(1.15(1.06-1.24);p<0.001),(1.01(1.00-1.02);p=0.012),(4.01(1.95-8.26);p<0.001),and(1.03(1.01-1.04);p<0.001),respectively.The multivariate analysis showed that the preoperative primary lesion SUVmax>6.3 or primary lesion size>2.3 cm were independent risk factors of peripheral cT1 NSCLC LNM(ORs,95%CIs were 6.18(2.40-15.92)and 3.03(1.35-6.81)).In this study,the sensitivity,specificity,PPV and NPV of the traditional 18F-FDG PET-CT diagnostic criterion(defined as hilar or mediastinal lymph node short diameter≥1.0 cm or lymph node SUVmax≥ 2.5)were 86.0%(37/43),39.7%(52/131),31.9%(37/116)and 89.7%(52/58),respectively.The AUC for predicting LNM with primary lesion SUVmax>6.3 or primary lesion size>2.3 cm was higher than that of traditional 18F-FDG PET-CT diagnostic criterion(0.687 vs.0.629;De-long test showed P=0.142),and its sensitivity and NPV was higher than that of traditional PET-CT diagnostic criterion(100.0%vs.86.0%,100.0%vs.89.7%).The results of 10-fold cross-validation showed that the mean values of AUC,sensitivity,specificity,PPV and NPV were 0.684,99.8%;37.0%,35.5%and 99.7%,respectively.A Hosmer-Lemeshow test showed a goodness-of-fit(P=0.479).Conclusions:The preoperative SUVmax>6.3 or the maximum diameter>2.3 cm of the primary lesions based on 18F-FDG PET-CT were the independent risk factors of peripheral cT1 NSCLC LNM.The sensitivity and NPV of preoperative SUVmax>6.3 or the maximum diameter>2.3 cm of the primary lesions were superior to the traditional 18F-FDG PET-CT diagnostic criterion,and it might identify the patients at low-risk LNM in peripheral cT1 NSCLC.Part Ⅲ Clinical Study on Predicting Lymph Node Metastasis with Radiomics in Primary Lesion based on 18F-FDG PET-CT of Non-small-cell Lung CancerObjective:At present,the new technology represented by artificial intelligence and radiomics is developing rapidly,and has been gradually applied in clinic.The machine learning model based on radiomics has certain clinical role in distinguishing benign and malignant lesions,predicting lymph node metastasis and evaluating therap eutic response and prognosis in NSCLC.However,there are few clinical studies on prediction of lymph node metastasis by establishing a nomogram combining 18F-FDG primary lesion metabolic parameters and machine learning radiomics models.and it is not clear whether the combination of primary lesion metabolic parameters and radiomics to establish nomogram has more advantages than simple machine learning radiomics model.The purpose of this study was to explore the best machine learning algorithm model for predicting lymph node metastasis in non-small-cell lung cancer in 18F-FDG PET-CT,and whether the nomogram established by combining 18F-FDG primary lesion metabolic parameters and radiomics machine learning model could accurately diagnose NSCLC lymph node metastasis preoperatively.Methods:The radiomics data and clinico-pathological information of patients with pulmonary lesions diagnosed by 18F-FDG PET-CT in the Department of Nuclear Medicine of the third affiliated Hospital of Soochow University from January 1,2017 to March 31,2021 were collected retrospectively.Take the data from January 1,2017 to March 31,2020 as the training set,and the data from April 1,2020 to March 31,2021 as the validation set.The primary lesions of all patients were segmented layer by layer manually by ITK-SNAP and 3D-Slices software to extract the imaging features of PET image and CT image.The extracted radiomics features included morphological features,first-order features,texture features and wavelet features.After preprocessing the extracted radiomics features,the best number of radiomics features was determined.The method was to set the number of features to 5,10,15,and 20,respectively,and the mean AUC was calculated by these seven machine learning methods:including random forest(RF),linear kernel support vector machine(SVMLinear),radial kernel support vector machine(SVMRadial),K nearest neighbor(KNN),neural network(NNET),adaboost(adaptive boosting)and Naive Bayes(NB).When the mean AUC was maximum,the corresponding number of features was the optimal number of radiomics features.According to the determined number of features,the maximum relevance and minimum redundancy method was used to screen the radiomic features,and the diagnostic performance of each machine learning algorithm model was compared,and these diagnostic performance indicators of each model were calculated respectively,including AUC,accuracy,specificity,sensitivity,PPV and NPV.Meanwhile,the RSD value of each machine learning algorithm model was calculated and determined the best ML model based on maximum AUC and minimum RSD.The best ML model was used to validate the validation set,and validated the model discrimination by calculating the ROC,and the confusion matrix was drawn to validate the model accuracy.Finally,three models(using only metabolic parameters,using radiomics only,combining metabolic parameters and radiomics)were established.And the ROC,calibration curves and decision curve analysis of the three models in training set and validation set were compared.And used the model with the largest AUC to build a nomogram.Results:A total of 3610 patients with pulmonary lesions were examined and diagnosed by 18F-FDG PET-CT in the Department of Nuclear Medicine of the third affiliated Hospital of Soochow University from January 1,2017 to March 31,2021.According to the inclusion and exclusion criteria,a total of 461 patients were included,including training group(n=308)and validation group(n=153).After comparing the mean AUC of different quantitative features,it was confirmed that the best number of radiomics features was 10(AUC=0.691).After compared and analied the diagnostic efficiency and RSD of seven machine learning algorithm models,it was found that the machine learning model based on the NB algorithm had the largest AUC and the smallest RSD.(AUC=0.718,RSD=0.129%).The results of univariate and multivariate logistic analysis showed that the SUR of the primary lesion was statistically different.The OR value and 95%CI of multivariate logistic analysis were 1.088(1.052-1.126),p<0.001.The nomogram was established by combining the SUR of the primary lesion and the radiomics features based on the NB model.The AUC in the training set and the validation set was 0.708(0.650-0.767),0.846(0.786-0.906),respectively,which was higher than that using only SUR models(training set:0.710(0.651-0.768),validation set:0.798(0.729-0.867))and models using only radiomics features(training set:0.728(0.67-0.786),validation set:0.815(0.746-0.884)).Conclusion:Combined the primary lesion SUR of 18F-FDG PET-CT standardized by mediastinal blood pool with the radiomics NB algorithm model to establish nomogram,the nomogram has higher diagnostic value and clinical efficacy in LNM of NSCLC,which can be used to help clinicians to choose surgical methods and provide basis for clinical decision-making information. |
PDF Full Text Request