Prediction of VISTA Expression and Its Prognostic Value in Pancreatic Cancer Using Contrast-Enhanced CT Radiomics

Pancreatic cancer remains highly lethal, and noninvasive biomarkers that capture the tumor immune microenvironment are urgently needed for individualized treatment and prognostic stratification. Here, we developed a contrast-enhanced computed tomography (CT) radiomics-based machine learning approach to preoperatively predict V-domain Ig suppressor of T-cell activation (VISTA) expression and evaluated its association with survival. In this retrospective single-center study, 89 patients who underwent surgical resection were analyzed using preoperative CT images, clinicopathologic variables, and survival data. VISTA expression from postoperative specimens served as the reference standard. Radiomic features were extracted from manually delineated tumor regions, selected using least absolute shrinkage and selection operator regression, and used to train 7 machine learning models with 10-fold cross-validation. Seven features were retained, and the random forest model achieved the best performance, with an AUC of 0.761. Low VISTA expression was independently associated with worse survival in multivariable Cox regression and improved time-dependent prognostic prediction when added to a baseline model. These findings suggest that CT radiomics may provide a noninvasive tool for predicting VISTA expression and refining prognostic stratification in pancreatic cancer.