Using CT image-based radiomics to predict EGFR mutations in lung adenocarcinoma

Abstract

Abstract OBJECTIVE To investigate the value of CT radiomics in predicting epidermal growth factor mutation status in lung cancer.MATERIALS AND METHODS Data from Hospital 1 (n = 119) and Hospital 2 (n = 49) were used as the internal dataset and data from the Cancer Genome Project (TCGA) database (n = 14) were used as the external validation set. After image segmentation, feature extraction and feature downscaling screening, radiomics models were constructed. Clinical features and conventional imaging markers used to predict EGFR mutation status were screened by univariate and multivariate logistic regression to construct the clinical model. Combining radiomic features and clinical features by multivariate logistic regression to construct a combined model. The predictive performance of each model was assessed using operator operating characteristic (ROC) curves, and the clinical utility of each model was assessed using clinical decision curves.RESULTS Three models based on CT flat-scan were developed: a radiomics model, a clinical model and a combined model. The area under the curve (AUC) for the training set radiomics, clinical and combination models were 0.0.87 (0. 79-0.95), 0.68 (0.57–0.80) and 0.87 (0.80–0.95) respectively; the AUC values for the validation set radiomics, clinical and combination models were 0.73 (0.55–0.90), 0.71 (0.54–0.88) and 0.73 (0.56–0.90) respectively.The clinical decision curves indicate that the combined model has better clinical decision benefits than the radiomics model alone and the clinical model alone.CONCLUSION The combined model combining radiomic features and clinical features further improved the predictive efficacy compared to the radiomic model alone and the clinical model alone.