BACKGROUND: In pancreatic surgery Postoperative pancreatic fistula (POPF) represents the most dreaded complication, for which pancreatic texture is acknowledged as one of the strongest predictors. No consensual objective reference has been defined to evaluate the pancreas composition. The presented study aimed to mine histology data of the pancreatic tissue composition with AI assist and correlate it with clinic-pathological parameters derived from the RECOPANC study. METHOD: From 320 patients originally included in the RECOPANC multicentric study, after series of exclusions slides of 134 patients were selected of AI-assisted analysis.For each slide tissue training fields were defined. Machine learning was trained to differentiate the tissue compartments: acinar, fibrotic, and adipose tissue, followed by quantification of the tissue area compartments. RESULTS: Relative fibrotic tissue area revealed as the strongest determinant for the prediction of clinically relevant POPF in multivariable analysis (p = 0.027). The AI assessed amount of fibrotic tissue performed significantly better in prediction of fistula development compared to the surgical palpatory assessment of the pancreatic texture. CONCLUSION: The present study is the first correlating AI-assisted quantified pancreatic tissue composition and POPF within a multicentric cohort.