Radiomics-Enhanced Hybrid Modeling of PDAC: A 3D Simulation Framework for Predicting Tumor Resectability

Pancreatic ductal adenocarcinoma is a highly aggressive malignancy with a poor prognosis, where accurate prediction of tumor resectability is critical for optimizing treatment strategies. Existing models for predicting resectability often fail to capture the complex biological interactions among tumor cells, vasculature, lymph nodes, and the tumor microenvironment, limiting their clinical utility. The proposed study presents a novel 3D hybrid modeling framework that integrates agent-based models and partial differential equations to simulate the dynamic evolution of tumor growth, vascular and lymphatic involvement, and microenvironmental interactions. The model incorporates critical clinical, radiomic, and anatomical features extracted from CT scans of PDAC patients, mapped onto a 3D voxelized grid aligned with patient-specific anatomy. Radiomic features are coupled with model parameters to capture spatial heterogeneity in tumor behavior. Model parameters are optimized using supervised machine learning techniques based on ground truth resectability labels provided by expert radiologists. Extensive simulations demonstrate the ability of the hybrid model to accurately predict tumor resectability status, achieving an accuracy of 99%, with corresponding precision, recall, and F1-scores of 0.99 across all classes. The model’s performance surpasses that of state-of-the-art classifiers, including XGBoost, DenseNet, Vision Transformers, and Graph Neural Networks, which achieved accuracies between 88% and 93%. These findings underscore the importance of integrating dynamic biological modeling with radiomics and clinical data for improved prediction of PDAC resectability.