Preoperative Prediction Of Lymph Node Metastasis In Non-Small Cell Lung Cancer Based On ¹⁸F-FDG PET/CT Radiomics

Objective:To explore the value of ¹⁸F-FDG PET/CT radiomics combined with clinical features and PET/CT parameters in preoperative prediction of mediastinal-portal lymph node metastases in non-small cell lung cancer（NSCLC）.Methods:Retrospective screening of 121 patients with NSCLC who underwent¹⁸F-FDG PET/CT with mediastinal-portal lymph node pathology findings at second hospital of Lanzhou university from March 2016 to February 2023,cases were randomly divided into a training set（n=85）and a test set（n=36）in a ratio of 7:3.In the training set,27 cases with lymph node metastasis and 58 cases without lymph node metastasis;15 cases of lymph node metastasis and 21 cases of no lymph node metastasis in the test set.Comparing the differences in clinical characteristics and PET/CT parameters between the lymph node metastasis group and the no lymph node metastasis group using univariate analysis,multivariate logistic regression analysis to screen independent predictors and construct clinical model.Extraction of radiomics features of primary lung cancer lesion based on ¹⁸F-FDG PET/CT images,application of LASSO regression for feature screening and logistic regression to develop separate PET,CT and PET/CT models.Calculation of radiomics score（RS）based on screened PET/CT radiomics features and development of a comprehensive model in conjunction with independent clinical predictors.The predictive performance of each model was assessed using receiver operating characteristic（ROC）curves and validated in the test set.Nomogram are drawn based on the best predictive performance model,calibration curve and decision curve analysis were used to assess the performance of the constructed models.Results:The differences in tumor markers（NSE,CY21.1,SCC,CEA）,short diameter,calcification,and PET/CT parameters of lymph nodes（SUVmax,SUVmean,SUVd,TLG,SUVratio）between the lymph node metastasis group and the group without lymph node metastasis were statistically significant（P<0.05）,with short diameter（OR=1.192,P=0.041）and SUVd（OR=1.208,P=0.035）of lymph node were independent predictors of lymph node metastasis.The area under the ROC curve（AUC）of the clinical model based on independent predictors were 0.793 and 0.743 in the training and test sets,respectively.The AUC of the PET,CT and PET/CT models based on radiomics features were 0.789,0.784 and 0.857 in the training set and 0.777,0.769and 0.828 in the test set,respectively.The AUC of the comprehensive model constructed by combining independent clinical predictors and RS was 0.917 and 0.873in the training and test sets,respectively,significantly better predictive performance than clinical model and other radiomics models.The decision curve analysis and nomogram demonstrate the value of the integrated predictive model for clinical applications.Conclusion:Compared to clinical or radiomics prediction models alone,the comprehensive model constructed by combining ¹⁸F-FDG PET/CT radiomics score and clinical independent predictors has better efficacy in predicting lymph node metastasis in NSCLC,corresponding nomogram can quantify the risk of lymph node metastasis prior to treatment.