Resection Margin Prediction in Pancreatic Ductal Adenocarcinoma: A Multi-Modal Machine Learning Approach Integrating Clinical and Computed Tomography Radiomics Data

We introduce an integrative machine learning model designed to improve the prediction of surgical margin status in patients with pancreatic ductal adenocarcinoma. Given the limitations of current predictive methods, we leverage both clinical and radiomics features across a retrospective dataset of 139 patients with available pre- and post-treatment CT imaging data. Our approach incorporates delta radiomics to quantify temporal changes in tumor morphology, providing a more dynamic perspective on disease progression. Our optimized model demonstrated superior classification performance compared to models based on single-feature groups alone. Additionally, we evaluated data availability across distinct time points to identify the most informative features for predicting surgical margin outcomes. Through extensive model fine-tuning, the integrative radiomics model demonstrated enhanced predictive capabilities, offering a promising tool to identify ideal surgical candidates and improve PDAC prognosis.