A Preliminary Study On The Prediction Of Mediastinal Lymph Node Metastasis In NSCLC Based On CT Radiomics Features

Objective:To investigate the predictive value of computed tomography(CT)radiomics features in the diagnosis of mediastinal lymph node(MLN)metastasis in patients with non-small cell lung cancer(NSCLC).Methods:110 patients with NSCLC who had mediastinal lymph nodes with a short diameter greater than or equal to 8 mm on CT were analyzed retrospectively,a total of 221 lymph nodes were included,of which 109 were positive(LN+)and 112 were negative(LN-),the dataset was randomly divided by the ratio of 7:3 into training and validation groups.A.K.software was used to delineate Region of interest(ROI)and extract radiomics features,the extracted features were analyzed by T test or Man-Whitney U test to eliminate features without statistical difference in LN-/+,the least absolute shrinkage and selection operator(LASSO)method was used for further feature screening to construct predictive models.The logistic regression classifier was used to establish the radiomics model in the training group,the radiomics score(Radscore)of each patient was calculated according to the regression coefficients of each parameter of the model.Univariate logistic regression analysis was used to screen the CT images and clinical pathological features with predictive value.Radscore as a label was then included in multivariate logistic regression analysis along with statistically significant CT images and clinical pathological features.Taking the minimum Akaike Information Criterion(AIC)as the criterion,stepwise regression analysis was used to further screen out the features with predictive value.Finally,a comprehensive model was established and expressed as radiomics nomogram.The performance of such model was analyzed by calibration curve and receiver operating characteristic curve(ROC)analysis in the training and validation group.Finally,the decision curve analysis(DCA)was conducted to evaluate the clinical usefulness of the prediction by quantifying the net benefits at different threshold probabilities.Results:The Area under the receiver operating characteristic curve(AUC)of the model established by radiomics features alone was 0.868 and 0.863 respectively in the training and validation group,while the AUC of the comprehensive model with CT images and clinical pathological features was up to 0.969 and 0.942,respectively(Delong test,p<0.05).Moreover,the application of radiomics nomogram in the training group and validation group had good predictive performance,and decision curve analysis showed that the multi-parameter radiomics nomogram had higher overall net benefit.Conclusion:The model constructed based on CT radiomics features can better predict the presence of MLN metastasis in NSCLC patients,while the multi-parameter radiomics nomogram combined with CT images and clinical pathological features can significantly improve its prediction efficiency and provide a reliable basis for personalized treatment of NSCLC patients.