AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-21
Towards reliable morphological subtyping and cancer detection on Barrett’s esophagus pathology through ensembling foundation models
The reliability of histopathology foundation models (Histo-FMs) for real-world diagnostic tasks has remained unproven. On the BarrettNet cohort of 3528 H&E-stained whole-slide images from 790 patients, spanning categories from normal to esophageal adenocarcinoma, the researchers tested their use for morphological subtyping of Barrett's esophagus and cancer detection. Aggregating the five most expert-consistent single-scale models across magnifications and backbones brought incremental gains (AUROC 0.907 for subtyping and 0.909 for cancer detection), while dysplasia assessment showed limited benefit and external validation remains necessary.