Deep learning system grades Barrett's esophagus dysplasia accurately across hospitals

Multi-institutional grading of Barrett's esophagus dysplasia without stain normalization using a pathology foundation model and multiple instance learning

Computer Vision and Pattern RecognitionArtificial Intelligence

Summary

Grading Barrett's esophagus dysplasia, a precancerous condition, can vary a lot between doctors, especially for mild cases, making treatment decisions hard. The authors developed a deep learning system that looks at microscope images and can classify the condition as normal, low-grade, or high-grade dysplasia. This system was tested on images from four different hospitals without needing to adjust for variations in how the samples were stained. It performed very well and consistently across all hospitals, helping reduce diagnostic uncertainty.

What this means in practice

  • For pathology labs: Implement automated grading systems that work accurately across multiple hospital sites without needing stain normalization.
  • For medical imaging software developers: Build pathology image analysis tools using foundation models robust to staining variations for broader clinical deployment.

Authors

Amirali Khosravi, Jaidip M. Jagtap, D. Chamil Codipilly, Catherine E. Hagen, Jason T. Lewis, Rickey E. Carter, Shahriar Faghani, Vani J. A. Konda, Fouad Otaki, Shajan Peter, Nicholas Shaheen, Sachin Wani, Prasad G. Iyer, Bradley J. Erickson

Abstract

Barrett's esophagus dysplasia grading suffers substantial inter-observer variability, particularly for low-grade dysplasia, complicating clinical decision-making. We conducted a retrospective multi-institutional diagnostic accuracy study of a foundation model-based deep learning system for dysplasia grading, using 969 H\&E-stained whole slide images from four institutions. Patch-level classifiers trained on expert annotations identified diagnostically relevant regions, followed by attribute-aware multiple instance learning for slide-level classification as non-dysplastic Barrett's esophagus (NDBE), low-grade dysplasia, or high-grade dysplasia. On a held-out test set (n = 130), macro-averaged AUC was 0.96 (95\% CI 0.93--0.98). For binary dysplasia detection, sensitivity was 98\% (80/82) and specificity 81\% (39/48), with a negative likelihood ratio of 0.030 and quadratic-weighted \kap{} of 0.83 against expert consensus. No high-grade dysplasia case was misclassified as NDBE. Performance was consistent across institutions without stain normalization, indicating that stain-robust foundation-model representations may remove the site-specific preprocessing that has constrained prior systems to institution-specific deployment.