An Iterative, Pathologist-in-the-Loop Workflow for Generation of Clinical-Grade Synthetic Pathology Images in a Diverse Cohort of Pancreatic Tumors

The training of diagnostic pancreatic pathologists is largely limited by the diversity of available pathology images. This study developed a workflow combining automated image processing with iterative feedback from expert pathologists to generate realistic, clinical-grade synthetic images of pancreatic tumors, with a focus on rare subtypes, to enrich educational cohorts. Quantitative evaluation showed that truncation increased precision while reducing recall and coverage, consistent with a quality-diversity trade-off. In an independent review blinded to image source and truncation condition, per-class truncation improved ratings of image quality and subtype representation relative to no truncation. After grouping ratings as 0–1 versus 2–3, the pathologists agreed on 80.7% of classifications, although agreement on the exact four-level score was lower. These results emphasize the importance of careful data selection, domain expertise, and independent multi-reader validation.