Development And Validation Of A CT Radiomics Model To Predict Preoperative Lymph Node Metastasis In NSCLC Patients

Objective: To establish a prediction model of lymph node metastasis(LNM)in non-small cell lung cancer(NSCLC)patients based on clinical and CT radiomics features and to verify the diagnostic efficiency of the model.Methods: A retrospective analysis of 90 patients with pathological diagnosis of NSCLC in our hospital.All patients underwent chest CT examination within two weeks before surgery and all lymph nodes had pathological reports.A total of 180 lymph nodes were included,including 78 metastatic lymph nodes(LN+)and 102 non-metastatic lymph nodes(LN-).The clinical data of patients were collected,including age,gender,pathological type of primary tumor,degree of differentiation,T stage.The imaging features of the patients were collected,including tumor location,maximum tumor diameter,lobulation sign,spiculation sign,pleural indentation sign,tracheal cutoff sign,vascular bundle sign,mediastinal and hilar enlarged lymph nodes,cavity or vacuolar sign.Import the patients’ chest CT image into 3D slicer software,and manually delineate the patient’s lymph nodes to obtain a region of interest(ROI).The delineated lymph nodes were randomly divided at a ratio of 7:3,and the radiomic features were extracted using the Pyradiomics software package.Predictive clinical factors were selected by univariate and multivariate logistic models.Radiomic features were screened by interclass correlation coefficient,single-factor analysis,least variance method,and the least absolute shrinkage selection operator(LASSO).Three prediction models(clinical model,radiomic model and clinical-radiomic model)were obtained.The performance of the model was evaluated using receiver operating characteristic(ROC)and Delong test.Use decision curve analysis(DCA)to quantify the net benefit value of various prediction methods when they reach a certain threshold,so as to evaluate the practicability of prediction models in clinical applicationResults: In the selection of clinical features,univariate and multivariate logistic regression analysis showed that the degree of differentiation(odds ratio [OR]: 29.803,95% confidence interval[95% CI]: 1.756-1430.602,P=0.049)and enlarged lymph nodes(OR: 4.474,95%CI: 1.455～14.792,P=0.01)was significantly correlated with LNM.All parameters were subjected to Akaike information criterion(AIC)multivariate logistic regression analysis and finally differentiated Three significant factors were the degree of lymph node enlargement and vascular cluster sign.In the selection of radiological features,through LASSO-logistic regression analysis,17 radiomics features(including 1first-order gray feature and 16 texture features)were finally found to be associated with lymph node metastasis in NSCLC.Based on the above characteristics,three models are finally established.The area under the curve(AUC)of the training group and the validation group were 0.784 and 0.585,respectively,the accuracy(ACC)was 0.754 and 0.574,and the sensitivity(SEN)was 0.8 and 0.783,respectively.The specificity(Specificity,SPE)were 0.718 and 0.419,respectively.In the radiomics model,the AUC values of the training and validation groups were 0.896 and 0.802,respectively,the accuracy was 0.81 and 0.759,the sensitivity was 0.945 and 0.913,and the specificity was 0.718 and 0.645,respectively.In the clinical-radiomic model,the AUC values of the training and validation groups were 0.928 and 0.815,respectively,the accuracy was 0.865 and 0.778,the sensitivity was 0.891 and 0.826,and the specificity was 0.845 and 0.742,respectively.The AUC values of the clinical model and the radiomics model,the training group and the validation group of the clinical model and the clinical-radiomics model were all statistically significant,and the P values of the Delong test were all <0.05.Analysis of the DCA results showed that the net benefit range curve of the radiomics model and the clinical-radiomics model in predicting LNM in patients with non-small cell lung cancer was higher than that of the clinical prediction model in both the training and validation groups.Conclusion: CT-based radiomics model and clinical-radiomics model have better predictive value for preoperative lymph node metastasis in NSCLC patients,and the predictive ability is better than clinical models.Compared with the radiomics model alone,the radiomics model with added clinical features has further improved diagnostic performance.