e20610 Background: Oligometastatic non-small cell lung cancer (NSCLC) exists high heterogeneity with distinct outcome, and there is a lack of available biomarkers for patient stratification. In this study, we identified a positron emission tomography (PET)/computed tomography(CT)-based radiomics signature capable of predicting overall survival (OS) in patients with synchronous oligometastatic NSCLC. Methods: This study consisted of 46 patients with synchronous oligometastatic NSCLC (≤5 metastases) between 2012-2018. Clinicopathologic data was acquired from medical records and database. A total of 20648 radiomic features were extracted from pretreatment CT and PET images, which were generated from the same PET/CT scanner. A radiomics signature was built by using the least absolute shrinkage and selection operator (LASSO) regression model. Multivariate Cox regression analysis was performed to establish the predictive model. The performance was evaluated with Harrell' concordance index (C-index). Results: 7 radiomics features were selected to build the radiomics signature. Multivariate analysis indicated that the radiomics signature (P = 0.007) was an independent prognostic factor, with a C-index of 0.810. Smoking status (P = 0.01) was the only independent clinicopathologic risk factor for overall survival prediction. Incorporating the radiomics signature with clinicopathologic risk factors resulted in higher performance with a C-index of 0.899. Conclusions: This study developed a radiomics model for predicting OS in synchronous oligometastatic NSCLC, which may serve as a predictive tool to identify individualized treatment strategy. Further internal and external validation of the model are required. Support: 81572279, 2016J004, LC2016PY016, 2018CR033. [Table: see text]
Overall survival prediction of non-small cell lung cancer by integrating microarray and clinical data with deep learning
