A spectral CT radiomics nomogram for preoperative discrimination between pancreatic neuroendocrine tumors and carcinomas

Abstract Objective To develop and validate a combined model integrating conventional CT, dual-layer spectral detector CT (DLCT), and radiomics features for preoperatively differentiating pancreatic neuroendocrine tumors (PNETs) from neuroendocrine carcinomas (PNECs). Materials and methods A total of 128 patients who underwent contrast-enhanced DLCT and were pathologically diagnosed with pancreatic neuroendocrine neoplasms (PNENs) were enrolled in this study. All patients were divided into PNETs group ( n = 86) and PNECs group ( n = 42) and randomly divided into training ( n = 89) and validation ( n = 39) groups in a 7:3 ratio. The general clinical data, conventional CT features and DLCT parameters of the patients were collected, and radiomics features were extracted based on 40 keV virtual monoenergetic images (VMIs) in venous phase. Logistic regression analysis was used to identify independent predictors for discriminating PNETs from PNECs. The conventional CT model, DLCT model, radiomics model based on 40 keV VMIs and combined model were constructed. The areas under the curve (AUCs) of each model were compared by Delong’s test, and the nomogram was plotted based on the optimal model. Results The combined model demonstrated superior predictive performance, achieving AUCs of 0.956 in the training cohort and 0.905 in the validation cohort. Delong’s test showed that in the training cohort, the combined model significantly outperformed the conventional CT model, DLCT model, and radiomics model (AUCs: 0.794, 0.835, and 0.871, respectively; all P < 0.05). In the validation cohort, the differences in AUC between the combined model and the conventional CT model, DLCT model, and radiomics model were not statistically significant (AUCs: 0.787, 0.805, and 0.811, respectively; all P > 0.05). Decision curve and calibration analyses further supported the robustness and favorable goodness-of-fit of the combined model in clinical application. Conclusion The combined model demonstrated promising performance in the identification of PNETs and PNECs, which may aid clinical decision-making, but its superiority requires further validation in external independent cohorts.