Radiomics model based on dual-energy CT venous phase parameters to predict Ki-67 levels in gastrointestinal stromal tumors.

Objective: To develop and validate a radiomics model based on the features of the Dual-Energy CT (DECT) venous phase iodine density maps and effective atomic number maps to predict Ki-67 expression levels in gastrointestinal stromal tumors (GISTs). Methods: A total of 91 patients with GIST were retrospectively analyzed, including 69 patients with low Ki-67 expression (≤5%) and 22 patients with high Ki-67 expression (>5%). Four clinical features (gender, age, maximum tumor diameter, and tumor location) were extracted to construct a clinical model. The venous phase enhanced CT iodine density maps and effective atomic number maps of DSCT were used to build radiomics models. Logistic regression was used to combine radiomics features with clinical features to build a combined model. Finally, the optimal model's discrimination, calibration, and clinical decision curve were validated using the Bootstrap method. Results: The combined model was identified as the best model, with high predictive performance. The model's discrimination had an AUC of 0.982 (95% CI, 0.9603-1). The calibration test showed a Hosmer-Lemeshow test P-value of 0.99. The clinical decision curve demonstrated a probability threshold range of 15% to 98%, with a high net benefit. Conclusion: The nomogram model combining clinical features and radiomics (iodine density map radscore + effective atomic number map radscore) has the highest accuracy for preoperative prediction of Ki-67 expression in GISTs.