Abstract
BACKGROUND: Accurate prediction of human epidermal growth factor receptor 2 (HER2) status is critical for personalized treatment strategies in gastric cancer (GC). This study aimed to develop and validate HER2 status prediction models integrating clinical and radiomics features to enhance predictive accuracy. METHODS: We retrospectively analyzed data from GC patients to identify independent predictors of HER2 status using multivariate logistic regression (LR). A radiomics model was constructed using five machine learning classifiers, with feature selection and Rad-score calculation. The combined model incorporated clinical predictors and Rad-score via LR. Model performance was evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). RESULTS: Pathological T stage, pathologic N stage, and CEA levels were identified as independent predictors of HER2 status (all p < 0.05). The support vector machine (SVM)-based radiomics model achieved the highest AUC (0.823) in the testing set. The combined model demonstrated superior predictive performance in both the training (AUC = 0.889) and testing (AUC = 0.826) cohorts compared to clinical model alone. CONCLUSIONS: The integration of CT-based radiomics with clinical factors significantly improved the prediction of HER2 status in GC, outperforming models based on either data type alone. TRIAL REGISTRATION: Not applicable (retrospective study).