In Non-Small Cell Lung Cancer, Can Radiomic Features Predict EGFR Mutations?

Abstract

Recently, targeted therapy for lung cancer epidermal growth factor receptor (EGFR) mutations has achieved significant clinical results. Therefore, it is very important to identify EGFR mutations for targeted therapy of non-small cell lung cancer(NSCLC).It is dangerous to get the surgery and biopsy samples for EGFR gene detection, in order to overcome this difficulty, we tried a non-invasive method using radiomic features extracted from medical images to predict EGFR mutations.170 patients with NSCLC who received surgery-based treatment were included in this retrospective study. CT scans were gathered and analyzed before invasive surgery. AK (Version V3.2.0.R) software was used to extract lesion features and feature selection. Then, a logistic regression model was established to predict EGFR mutation status. Finally, diagnostic performance of characteristic parameters was analyzed according to the receiver operating characteristic (ROC) curves. Therefore, the radiomic features with higher diagnostic value were screened out. In addition, we used univariate analysis to analyse the association between clinical features and the presence of EGFR mutations. Radiomic features can construct predictive models of EGFR mutations. The area under the receiver operating characteristic curve (AUC) of the training group was 0.766, and the AUC of the testing group was 0.748, both of which at good diagnostic decision point. There is a correlation between clinical features and EGFR mutations, and predictive models can also be established. EGFR mutation status can be predicted well by radiomic features. Predictive models can also be constructed based on clinical features. The combination of clinical features and radiomic features will have higher diagnostic value.