A CT-Based Radiomics Model for Preoperative Risk Stratification of Gastrointestinal Stromal Tumors

Tianyi Wan¹Yanping Song²Xiaolian Wang¹Huang Pei¹Zonghuo Wang¹Bing Fan^1*Wentao Dong^1*

• ¹Jiangxi Provincial People's Hospital, Nanchang, China
• ²The First Affiliated Hospital of Nanchang University, Nanchang, China

Background: Gastrointestinal stromal tumors (GISTs) are common mesenchymal tumors with variable malignancy potential, making accurate preoperative risk stratification crucial for treatment planning. Traditional methods rely on pathological and clinical features but often overlook tumor heterogeneity. This study aims to develop and validate a CT-based radiomics model for GISTs risk stratification to improve clinical decision-making. Methods: This retrospective, multi-center study developed and validated a radiomics-based risk prediction model in accordance with the TRIPOD-ML statement. It included 123 patients with GISTs from two hospitals, divided into training (n=68), testing (n=30), and external validation (n=25) cohorts. Tumor delineation was performed using 3D segmentation on venous-phase contrast-enhanced CT scans. Radiomics features (n=1784) were extracted and refined using feature selection methods, including LASSO and ANOVA. Six machine learning algorithms were evaluated, and the support vector machine (SVM) model demonstrated optimal performance. Model evaluation included metrics such as AUC, calibration curves, and decision curve analysis (DCA). Results: The SVM-based radiomics model achieved robust performance, with AUC values of 0.906 (95% CI: 0.812–0.964) in the testing cohort and 0.867 (95% CI: 0.724–0.956) in the external validation cohort. Calibration curves indicated strong agreement between predicted and observed outcomes, while DCA highlighted significant clinical utility across different thresholds. Key radiomics features provided accurate differentiation between Lower Risk and Elevated Risk groups, aligning with clinical stratification needs. Conclusions: The developed CT-based radiomics model offers a reliable, externally validated tool for GISTs risk stratification, addressing limitations of traditional methods by incorporating tumor heterogeneity and enhancing predictive accuracy. This model has the potential to guide personalized treatment strategies, particularly in distinguishing patients with GISTs requiring adjuvant therapy from those suitable for surgical resection alone. This study was approved by the appropriate ethics committee with a waiver of informed consent.