AI-powered analysis of Pancreatic Ductal Adenocarcinoma tissues to study the tumor immune ecosystem and identify novel classifiers

Background: Digital pathology and artificial intelligence (AI) are emerging as powerful tools in immuno-oncology, enabling enhanced diagnostic and prognostic workflows, by extracting features from tumor slides beyond visual perception. Pancreatic ductal adenocarcinoma (PDAC), one of the deadliest cancers globally, lacks effective biomarkers for patient stratification and therapy response assessment. Aim: This study presents a computational pipeline to integrate the analysis of PDAC tissues stained with different methods and identify features with clinical relevance. Methods: We analyzed whole slide images (WSIs) from 55 PDAC patients, both treated and untreated with neoadjuvant chemotherapy. Slides were stained with H&E and CD68 to detect macrophages. A pixel classifier in QuPath segmented CD68+ areas, enabling calculation of immune-related area percentage (IRA%) and Morisita Index. A pre-trained tissue classifier analyzed H&E slides to quantify regions (Tumor, Stroma, Lymphocytes, etc.) and compute spatial tissue entropy. Results: We found that higher CD68+ IRA% and Morisita Index were significantly associated with worse overall survival. Increased Tumor Entropy also correlated with poorer prognosis. Notably, combining Morisita Index with Tumor Entropy further improved patient survival stratification. We also observed differences in these features between chemotherapy-treated and untreated patients. Conclusions: Our AI-driven pipeline enables quantitative and spatial profiling of tumor and immune microenvironments in PDAC, identifying interpretable features with prognostic value. These findings support the development of novel, clinically relevant biomarkers through computational pathology.