Abstract
PURPOSE: To develop and validate a CT-based radiomics model for differentiating primary gastric lymphoma (PGL) from Borrmann type IV gastric cancer (GC). MATERIALS AND METHODS: A total of 136 patients with pathologically confirmed PGL (n = 56) and Borrmann type IV GC (n = 80) were retrospectively enrolled between January 2016 and May 2022. The cohort was randomly partitioned into a training set (n = 95) and a testing set (n = 41) at a 7:3 ratio. Radiomics features were extracted from unenhanced, arterial, venous, double-phase (arterial + venous), and three-phase (unenhanced + arterial + venous) CT images. After feature selection using the Least Absolute Shrinkage and Selection Operator, radiomics models were constructed via logistic regression. A clinical-radiomics model was developed through multivariate analysis. The models were evaluated using Receiver Operating Characteristic (ROC) curves, calibration curves with the Hosmer-Lemeshow test, and decision Curve Analysis (DCA) for clinical net benefit. RESULTS: Clinical model comprised of high-enhanced serosa sign, normalized CT value on venous phase, and perigastric fat infiltration showed good performance with AUCs of 0.902 (training set) and 0.878 (testing set). Among the radiomics models, the three-phase model outperformed others (AUC: 0.871 training, 0.865 testing). The clinical-radiomics combined model further improved discriminatory performance, achieving AUCs of 0.960 and 0.932 in the training and testing sets, respectively. DCA confirmed that the combined model provided the highest clinical net benefit. CONCLUSION: Clinical-radiomics model incorporating three-phase radiomics signatures and CT findings achieved satisfactory performance for differentiating PGL from Borrmann type IV GC, serving as a reliable non-invasive tool for clinical decision-making.