Abstract Objective To develop and validate a CT-based radiomics model to predict immunotherapy response in unresectable gastric cancer and explore its underlying biological mechanisms. Materials and methods This retrospective study included 368 unresectable gastric cancer patients receiving programmed death-1/programmed death ligand-1 (PD-1/PD-L1) inhibitors combined with chemotherapy from two centers. Patients were divided into training ( n = 231), internal validation ( n = 97), and external validation ( n = 40) cohorts. Radiomics model was constructed using portal venous phase CT images, and a radiomics score (Radscore) was calculated for each patient. Five machine learning models incorporating clinical factors and Radscore were developed and compared. The best-performing model was used to construct a nomogram. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Immune cell infiltration analysis was performed using data from The Cancer Genome Atlas (TCGA) cohort. Results The radiomics signature, comprising 15 selected features, showed good predictive performance across all cohorts: training (AUC = 0.868), internal validation (AUC = 0.816), and external validation (AUC = 0.793). The logistic regression model demonstrated the highest and most consistent performance, with AUC values of 0.886, 0.831, and 0.826, respectively. The developed nomogram showed satisfactory calibration and clinical utility. Immune infiltration analysis revealed significant associations between Radscore and infiltration levels of activated CD4+ memory T cells, regulatory T cells, and CD8+ T cells. Conclusions The CT-based radiomics nomogram showed promise for personalizing immunotherapy treatment strategies in unresectable gastric cancer. The association between the Radscore and immune cell infiltration provided insights into its biological basis. Critical relevance statement This rigorously validated CT radiomics nomogram critically advances gastric cancer immunotherapy prediction, offering clinical radiology a non-invasive, biologically-informed tool to guide personalized treatment decisions. Key Points CT radiomics provided a reliable marker for predicting gastric cancer immunotherapy response. The developed Radscore correlated with immune cell infiltration, offering biological insights. A nomogram integrating the Radscore and clinical factors showed robust predictive performance. Graphical Abstract
Zhan et al. (Thu,) studied this question.