A nomogram to preoperatively predict the aggressiveness of pancreatic neuroendocrine tumors based on CT features and 3D CT radiomic features.

Objectives: Combining Computed Tomography (CT) intuitive anatomical features with Three-Dimensional (3D) CT multimodal radiomic imaging features to construct a model for assessing the aggressiveness of pancreatic neuroendocrine tumors (pNETs) prior to surgery.

Methods: This study involved 242 patients, randomly assigned to training (170) and validation (72) cohorts. Preoperative CT and 3D CT radiomic features were used to develop a model predicting pNETs aggressiveness. The aggressiveness of pNETs was characterized by a combination of factors including G3 grade, nodal involvement (N + status), presence of distant metastases, and/or recurrence of the disease.

Results: Three distinct predictive models were constructed to evaluate the aggressiveness of pNETs using CT features, 3D CT radiomic features, and their combination. The combined model demonstrated the greatest predictive accuracy and clinical applicability in both the training and validation sets (AUCs (95% CIs) = 0.93 (0.90-0.97) and 0.89 (0.79-0.98), respectively). Subsequently, a nomogram was developed using the features from the combined model, displaying strong alignment between actual observations and predictions as indicated by the calibration curves. Using a nomogram score of 86.06, patients were classified into high- and low-aggressiveness groups, with the high-aggressiveness group demonstrating poorer overall survival and shorter disease-free survival.

Conclusion: This study presents a combined model incorporating CT and 3D CT radiomic features, which accurately predicts the aggressiveness of PNETs preoperatively.