Prediction And Risk Stratification Of Distant Metastasis Of Non-small Cell Lung Cancer Based On CT Radiomics

Background and purpose:Lung cancer is a highly malignant tumor with high morbidity and mortality worldwide.In my country,the morbidity and mortality of lung cancer have increased significantly in recent years.Non-small cell lung cancer(NSCLC)is the main type of lung cancer,accounting for about 80% to 85% of the total.Most patients with early lung cancer have no clinical symptoms,and most of them are in the late stage when symptoms appear.The clinical treatment effect is poor,the prognosis is poor,and the 5-year survival rate is low.The stage of lung cancer has important guiding significance for the formulation of reasonable treatment plans and prognosis assessment for lung cancer patients.The 5-year survival rate of patients with lymph node metastasis(LNM)and distant metastasis(DM)will be greatly reduced.The heterogeneity of the tumor itself and the micro-infiltration around the tumor are the key factors to determine whether DM occurs.For the micro-infiltration around the tumor,the macroscopic morphological analysis of imaging has certain limitations,and the microscopic pathological observation can achieve some effect.However,the acquisition of pathological specimens is invasive and lagging,which limits its large-scale application.Radiomics is a new technology developed in the past ten years.Through high-throughput data extraction from traditional imaging images and correlation with the cellular and molecular mechanisms of lesions,non-invasive detection of lesions can be achieved,and radiomics technology is analyze the image data,the results can be obtained in a timely manner.Due to the heterogeneity of the tumor itself,the risk of DM in the tumor is also different.Therefore,early and timely prediction of the risk of DM in NSCLC patients and stratification of the risk level,it will be of great importance guiding significance for different clinical interventions.The purpose of this study is to use radiomics technology to analyze computed tomography(CT)images of NSCLC patients before treatment,extract the radiomics features inside and around the tumor.Then construct a clinicopathological risk factor model,a tumor radiomics model,a peritumoral radiomics model and combined model were used to evaluate their ability to predict DM and risk stratification in NSCLC patients,and to evaluate their distant-metastasis-free survival(DMFS).Materials and methods:In the First Affiliated Hospital of Army Medical University,140 patients with NSCLC were retrospectively analyzed,including 74 patients without DM and 66 patients with DM.Inclusion criteria for patients were: 1)pathological type of NSCLC;2)no metastasis before treatment;3)chest CT examination within 1 month before treatment;4)non-surgical treatment.The exclusion criteria were: 1)combined with tumors of other systems;2)incomplete clinical and pathological follow-up data;3)poor quality of CT images or unrecognizable tumor boundaries.Chest CT images of each patient before treatment were collected,and two regions of interest(ROI)were delineated layer by layer from the thin layer of the lung window,including the primary tumor area(denoted as tumor)and the peritumoral micro-invasion region(denoted as ME)that expanded uniformly outward with 10 mm,pay attention to the elimination of large vessels,bronchi and bone in the peritumoral area.Then the radiomics features were extracted and the radiomics score(RS)was calculated.The consistency of feature delineation was tested by the intraclass correlation coefficient(ICC).All data were randomly divided into groups according to 7:3,and there were 98 cases in the training cohort and 42 cases in the verification cohort.In the process of constructing the DM model,variance threshold method,univariate selection method and the least absolute shrinkage and selection operator(LASSO)regression method were used for feature selection.In terms of model construction,four prediction models were constructed,which were tumor radiomics model,peritumoral(tumor+ME)radiomics model,and their respective integrated models combined with clinical features.At the same time,a nomogram of a comprehensive model containing peritumoral(tumor+ME)radiomics combined with clinical features was constructed.The performance of all models was quantified by the area under the curve(AUC)of the receiver operating characteristic(ROC)curve,and the diagnostic ability of different models was tested for significance using the De Long test.In the process of constructing the DMFS prognostic model,univariate and multivariate COX regression analysis were used to screen out the clinicopathological risk factors significantly related to DMFS.The LASSO-COX regression method was used to reduce the redundancy of features and select the most useful prognostic features related to DMFS.In the aspect of model construction,the clinicopathological risk factor model,RS model and clinical-RS comprehensive model were developed,and the clinical-RS comprehensive model nomogram was constructed for the visualization of DMFS.In the training cohort,the difference in DMFS between high and low risk scores was analyzed using Kaplan-Meier survival curves and log-rank tests,and the obtained results were applied to the validation cohort.Results:Among the clinicopathological risk factors,carcinoembryonic antigen(CEA)had statistical significance in the prediction of DM in NSCLC(P<0.05).In the DM model,the AUC values of the radiomics models of the tumor group and the peritumoral group were0.779 and 0.854 in the training cohort,and 0.648 and 0.804 in the validation cohort;In the models,the AUC values of the tumor group and the peritumoral group in the comprehensive models after adding clinicopathological risk factor in the training cohort were 0.795 and 0.858,and the AUC values in the validation cohort were 0.674 and 0.828,respectively.A model including peritumoral radiomics features combined with clinicopathological risk factors has the best diagnostic performance in predicting DM in NSCLC.The De Long test results of the peritumoral radiomics combined with the clinical comprehensive model and the other three models showed that there were statistically significant differences between the model with peritumor and the model without peritumor in both the training cohort and the validation cohort(P<0.05).In the DMFS prognosis model,the clinical-RS comprehensive model showed the best predictive efficiency,and its consistency index(C-Index)in the training cohort was 0.730,95%confidence interval(CI),0.689-0.771,and the C-Index of the verification cohort was 0.722,95% CI,0.660-0784.The C-Index of the single clinicopathological risk factor model was0.719 in the training cohort and 0.642 in the verification cohort,while the C-Index of the RS model was 0.676 in the training cohort and 0.642 in the verification cohort.The patients were divided into high and low risk group by KM curve,and the risk score was significantly correlated with the decrease of DMFS in the training cohort(p < 0.0001).Conclusions:The radiomics features can predict DM in NSCLC,and a model including peritumoral radiomics can improve the prediction performance.The clinical-radiomics comprehensive scoring model can non-invasively predict the DMFS of NSCLC and realize its risk stratification,which can provide certain help for clinical treatment selection and follow-up.