CT-based Radiomics Features To Predict The Survival Outcomes And Imaging Subtype Of Non-Small Cell Lung Cancer

According to the National Cancer Center,lung cancer remains the leading cause of morbidity and mortality among all malignant tumors,and non-small cell lung cancer(NSCLC)accounts for 80%-85%of all lung cancer cases.The growth and division of cancer cells in NSCLC is slower and the diffusion and metastasis are relatively late.However,due to the outbreak of the disease,there is a lack of specific symptoms in the early stages of the disease,about 80%of lung cancer patients are in the middle and late stage of diagnosis,so that this disease is accompanied by a very low survival rate.The occurrence and progression of lung cancer is a multi-factorial participatory process,the complexity of which not only reflected in the genetic heterogeneity of the tumor,but also in the spatial heterogeneity of the tumor.The development of radiomics and genomics provides a new way to understand the heterogeneity of tumors.Using imaging methods to assess the heterogeneity of tumor heterogeneity,and then correlate its differences in genetic expression,which not only to providing a reference for optimizing clinical treatment strategies,but also promote the implementation of individualized precision treatment and improve the prognosis survivability of patients.In this paper,we retrospectively analyzed the NSCLC data in the TCIA and TCGA databases and 371 patient were included.The collection of NSCLC-Radiomics contains CT images and clinical data,and was divided into training cohort and validation cohort,the external validation cohort acquired from the collection of TCGA-LUSC/LUAD,which contains patient imaging,clinical and gene expression data.The semi-automatic segmentation of region of interest(ROI)performed using the 3D-slicer,the 3D-labeling ROI covered the whole tumor area and the 2D-labeling ROI delineated the maximum cross-sectional area of a tumor.975 radiomics feature were extracted from each patient’s 2D and 3D CT images,including shape features,first-order statistics feature,texture features,and wavelet features.Firstly,we evaluated the predictive performance of 2D and 3D radiomics features on patient survival time using univariate Cox regression model and the K-M survival analysis method,and analyze the consistency of radiomics features between different dimensions.Then,multivariate Cox proportional hazard model used to determine the association between radiomics signature and the survival probability of patients adjusted for other clinical variables.Finally,we used unsupervised consensus clustering to discover and validate the intrinsic imaging subtypes of non-small cell lung cancer.We evaluated the imaging subtypes in terms of their prognostic capacity for predicting survival time and performed pathway analyses to elucidate the biological mechanism of the imaging subtypes.We found both 2D and 3D radiomics signatures of the tumor have favorable prognosis,but the 3D radiomics signature had a better performance.The radiomics signature generated from the combined 2D and 3D image features had a best predictive performance than those from 2D or 3D features.Compared with the prognostic model using clinical information alone,the prognostic model integrating the optimal radiomics signature with clinical predictors significantly improved the predictive performance in the survival probability of patients.We have identified three lung CT imaging subtypes based on quantitative imaging phenotypes of the tumor.The three imaging subtypes reflect distinct underlying molecular pathways,and are associated with significantly different survival.The analysis of gene enrichment pathway shows that the imaging subtypes were correlated with the degree of misalignment of signaling pathway,and the survival probability of patients’ decreases significantly with the increase of signal pathway disorder and pathway activity.The results shown that the radiomics can achieve a comprehensive description of tumor spatial heterogeneity,which not only provide comprehensive information for clinical treatment,but also reveal the relationship between imaging and genetic heterogeneity,to provide reference for the accurate treatment of non-small cell lung cancer.