AI model reveals long-hidden pancreatic tissue changes in type 2 diabetes

Deep learning analysis of high-resolution pancreatic tissue images has identified subtle structural changes linked to beta cell dysfunction which offers novel insight into the development of type 2 diabetes

Researchers from several partner institutions of the German Center for Diabetes Research have collaborated with international colleagues to develop a novel approach to visualise subtle tissue alterations in the pancreas associated with type 2 diabetes (T2DM). The findings have provided important insight into disease mechanisms that conventional pathology has struggled to detect.

More than 500 million people worldwide live with T2DM, a chronic metabolic disorder characterised by impaired insulin secretion and reduced insulin sensitivity. Many patients experience serious complications that affect the cardiovascular system, kidneys, eyes and peripheral nerves.

Despite the global burden of disease, classical histopathological examination of pancreatic tissue has rarely enabled clinicians to draw robust conclusions about an individual’s glycaemic status. Subtle morphological alterations that accompany beta cell dysfunction have often remained almost imperceptible under standard microscopic assessment.

To address this diagnostic limitation, the research team assembled an extensive dataset derived from pancreatic tissue sections obtained from living donors. The investigators applied chromogenic staining which relies on enzyme-mediated colour development, alongside multiplex immunofluorescent staining to label multiple cellular markers within a single section. They then captured the stained samples at ultra-high resolution through gigapixel microscopy, a technique that allows detailed visualisation of large tissue areas without loss of cellular detail.

On this foundation, the scientists trained deep learning models to analyse complex image patterns beyond the threshold of human perception. These models distinguished reliably between tissue samples from individuals with and without T2DM. The system did not simply classify samples but also identified structural features that contributed most strongly to its predictions.

The analysis revealed that specific alterations within the islets of Langerhans, which contain the insulin-producing beta cells and glucagon-secreting alpha cells, played a central role. The models also detected changes in neuronal axons within the pancreatic microenvironment and highlighted the spatial proximity of adipocyte clusters to islet structures. Such architectural rearrangements have long been suspected to influence metabolic regulation, yet conventional microscopy has rarely enabled systematic quantification.

To ensure transparency and clinical interpretability, the team employed explainable artificial intelligence (AI) methods. These approaches enabled researchers to trace which image features influenced the algorithm’s decisions and to quantify them as candidate biomarkers. Rather than operate as an opaque ‘black box’, the system provided a structured map of disease-associated tissue features that clinicians and pathologists can evaluate.

This AI-supported evaluation has therefore delivered insight into early and previously difficult-to-detect pancreatic changes in T2DM. By combine high-resolution imaging with computational pattern recognition, the study has demonstrated how digital pathology can uncover biological signals that escape routine examination.

The findings have opened new perspectives on the processes that unfold in the pancreas during the development of T2DM. Improved characterisation of microstructural alterations may, in time, refine diagnostic stratification and support more precise therapeutic intervention.

For further reading please visit: 10.1038/s41467-026-69295-2