A team from the German Center for Diabetes Research trained explainable deep‑learning models on gigapixel microscopy of pancreatic tissue from living donors and identified microscopic features associated with impaired insulin secretion and beta‑cell dysfunction. The models reliably distinguished diabetic from non‑diabetic samples and highlighted alterations in islets and non‑beta cell populations. By using living‑donor tissue and explainability methods, the study exposes subtle histopathological signals that conventional pathology misses and provides candidate tissue biomarkers for early type‑2 diabetes biology.