Bronchial artery chemoembolization (BACE) is regarded as a safe and effective treatment method for advanced lung cancer. However, the therapeutic effects of BACE vary greatly, and there is no reliable prognostic tool in clinical practice.The aim of this study was to develop and validate a model based on computed tomography (CT) radiomics for predicting tumor response to bronchial artery chemoembolization (BACE) in advanced lung cancer (Stage III–IV) after failure of first-line therapy. A total of 227 patients from three centers enrolled this retrospective study. Radiomic features were derived from arterial phase CT images, and feature selection was performed successively using variance thresholding, univariate feature selection, and least absolute shrinkage and selection operator (LASSO) regression. Five machine learning classifiers were employed to calculate radiomics (Rad)-scores. A fusion model was developed based on the Rad-scores and independent clinical predictors. Five important radiomics features were ultimately identified and used to create the models. The LightGBM model had the highest efficiency, with an area under the curve (AUC) of 0.809 and 0.746 for the internal validation and external validation cohorts, respectively. The LightGBM-based Rad-score was combined with independent clinical predictors (ECOG Score, blood supply count, and ProGRP) to generate the fusion model, which achieved better predictive performance (AUC = 0.928, 0.875, and 0.813 in the training, internal validation, and external validation cohort, respectively) than the radiomics model (AUC = 0.872, 0.809, and 0.746, respectively) and clinical model alone (AUC = 0.750, 0.732, and 0.672, respectively). The fusion model could effectively predict the tumor response to BACE in lung cancer and help clinicians identify the appropriate surgical population.
Yang et al. (2026) studied this question.