Preoperative prediction of MSI status in gastric cancer based on CT and radiomics: a multicenter study

To develop a non-invasive prediction model and assess its predictive power of microsatellite instability (MSI) status in gastric cancer (GC). This retrospective study included 1,012 GC patients from three hospitals during January 2019 and January 2023. Patients from hospital C were independently designated as the external test cohort. MSI status was determined by immunohistochemistry. Radiomic features were extracted from delineated tumor regions on preoperative venous phase CT images to develop radiomic signatures. The predictive performance was evaluated by Area Under the Curve (AUC), sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV). Violin plots illustrated the distribution of rad-score. Multivariable logistic regression was conducted to establish a nomogram combining two clinical features and rad-score. The model demonstrated clinical applicability by decision curve, and impact curve. Seventy-four of 1012 (7.31%) patients were MSI-High. The radiomic model (Z-score_PCC_RFE_LRLasso) with 16 features, yielded an AUC of 0.805 (95% CI: 0.753, 0.856) in training cohort and 0.772 (95% CI: 0.690, 0.831) in test cohort. The nomogram showed improved risk reclassification and comparable discriminative performance (AUC, 0.836; 95% CI: 0.788, 0.885; AUC, 0.808; 95% CI: 0.731, 0.884, respectively). Its NPV and specificity was 0.959 (95% CI: 0.927, 0.980) and 0.800 (95% CI: 0.750, 0.844). Radiomics and hybrid nomogram have potential in predicting MSI status in GC. This non-invasive diagnostic model can be further refined.