Computed Tomography-Based Delta-Radiomics for Early Prediction of Short-Term Response to Concurrent Chemoradiotherapy in Patients with NSCLC

Abstract

This study aimed to investigate the possibility of CT-based delta-radiomics for early prediction of short-term response in non-small cell lung cancer (NSCLC) during concurrent chemoradiotherapy and determine the optimal time point for prediction of short-term response. Twenty patients with NSCLC who received concurrent chemoradiotherapy were prospectively enrolled. A total of 1210 delta-radiomic features (ΔRFs) were calculated from each planning and follow-up three weekly CTs per patient. The feature selection for ΔRFs was performed using intra-class correlation coefficient (ICC), Pearson correlation, ANOVA-test or Mann-Whitney U-test, and univariate logistic regression. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve was preliminarily used to evaluate the prediction ability of short-term responses (first and third months) at different time points (P < 0.05). Of the 1210 ΔRFs for weeks 1-3, 121 common features were retained after ICC and Pearson's correlation. Subsequently, 54 and 58 features at all time points were significantly different between the response group and non-response group for the first and third months, respectively (P< 0.05). Subsequently, 11 and 44 features remained after univariate logistic regression for the first and third months, respectively. Finally, eight ΔRFs that were able to discriminate short-term responses at both the first and third months with statistical accuracy were identified. CT-based delta-radiomics may potentially provide reasonable biomarkers of short-term response to concurrent chemoradiotherapy for NSCLC, which can help improve clinical decisions for early treatment adaptation.