Exploring The Value Of CT1 Stage NSCLC Radiomic Features To Predict Mediastinal Lymph Node Metastasis

Objective:Investigate the value of predicting mediastinal lymph node metastasis based on CT-based radiomics biomarkers and clinical features in cT1 nonsmall cell lung cancer(NSCLC).Method:From August 2013 to October 2019,there are 196 patients in stage T1 non-small cell lung cancer were selected from the lung cancer patients who underwent surgical cut off in BethuneHospital of Jilin University,and their clinical features、pathological datas were analyzed.The clinical datas include age、gender,tumor pathological information includes tumor histology type,peritumoral changes(include vascular invasion,nerve infiltration,chest wall invasion,small bronchial wall infiltration,trachea and vascular margins see cancer).every patient underwent a CT scan of chest before surgery to collect CT image features(size,side,location),Segment the ROI(region of interest)of the primary lesion and extract radiological features.Then the dimension has been reduced through a specific algorithm which called LASSO and then filter out the features with more diagnostic value.Using Support Vector Machine classifier to build a predictive model of pure radiomic features,a prediction model combined clinical features with radiology characteristics.Receiver Operating Characteristic Curve(ROC)was bewrited by taking the 5-fold cycle assessment method,and applying the Area Under Curve(AUC)and 95% confidence interval(CI)were to considerated the two predictions Model predicts diagnostic efficacy of preoperative lymph node metastases.Results:Among the clinical variables,gender,tumor location,histological subtype,and peritumoral changes(vascular invasion,chest wall infiltration,and nerve infiltration)were correlation arguments of lymph node metastasis in sick persons in cT1 stage NSCLC(P <0.05).Each ROI can extract 1289 radiomic features,and after dimensionality reduction by LASSO,23 radiomic signatures can be obtained.The AUC calculated by the pure radiomic features prediction model in the training cohort,test ground and validation cohort are 0.95 ± 0.00,0.806,0.77 ± 0.08 respectively,95% confidence intervals were 0.683-0.929、0.883-0.982;The AUC of the predictive model combined by radiomic features and clinical features were 0.94 ± 0.01,0.864,0.91 ± 0.05,and the 95% confidence intervals were0.892-0.981、0.763-0.964.Conclusion:Preoperative lymph node metastasis prediction model constructed by radiomic features,gender,lesion location,histological subtype,nerve invasion,vascular invasion,and chest wall invasion has better diagnostic performance than the radiomic features prediction model alone.It is of great significance for the selection of individualized treatment for different patients to determine the preoperative lymph node stage in patients with non-small cell lung cancer.