Integration of dual‐energy CT characteristics and biomarkers: Noninvasive prediction of Ki‐67 expression in pancreatic ductal adenocarcinoma

BACKGROUND: The Ki-67 proliferation index is a critical prognostic marker in pancreatic ductal adenocarcinoma (PDAC); however, its assessment relies on invasive tissue sampling. Ki-67 expression reflects active tumor cell proliferation and is associated with aggressive tumor behavior. A preoperative, noninvasive method to predict Ki-67 status would therefore be valuable for clinical decision-making. Dual-energy CT (DECT) can provide quantitative parameters related to tumor vascularity and composition, potentially reflecting proliferative activity. Additionally, clinical biomarkers such as CA125 may offer complementary information regarding tumor biology. Therefore, the development of a reliable noninvasive approach to preoperatively determine Ki-67 status is of considerable clinical importance. PURPOSE: To develop and validate a noninvasive approach for predicting Ki-67 expression in pancreatic ductal adenocarcinoma by integrating quantitative dual-energy CT parameters and clinical biomarkers. METHODS: This retrospective study included 148 PDAC patients randomly divided into training (n = 89) and validation (n = 59) sets (6:4 ratio). All patients underwent preoperative DECT scans, and quantitative parameters including normalized iodine concentration (NIC), effective atomic number (Zeff), spectral attenuation slope (λ), etc. were obtained from three contrast phases. Serum tumor markers (CA19-9, CA125, CA50, CEA) and clinical features were analyzed. Multivariate logistic regression was used to identify predictors of Ki-67 expression. A nomogram and 3-D probability surface were developed to intuitively demonstrate the model's predictive structure and decision-making process. Model performance was validated using ROC analysis, calibration curves, and decision curve analysis. Innovatively, kernel-density ridgeline plots and prediction-error bar plots were employed to comprehensively evaluate risk distribution and prediction accuracy, demonstrating the model's stability. RESULTS: The joint model demonstrated excellent predictive performance, achieving AUCs of 0.803 in the training set and 0.810 in the validation set, outperforming both the clinical-only model (training AUC = 0.682, validation AUC = 0.751) and the DECT-only model (training AUC = 0.712, validation AUC = 0.702). Multivariate analysis identified arterial-phase normalized iodine concentration (A-NIC) (p = 0.046) and CA125 (p = 0.005) as independent predictors of Ki-67 expression. These two parameters formed the basis of the final predictive model, demonstrating consistent diagnostic value across both cohorts. CONCLUSION: Integration of DECT parameters and clinical biomarkers allows accurate noninvasive prediction of Ki-67 expression in PDAC, offering a potential tool for preoperative assessment of tumor proliferation.