Discriminative Self-Supervised Pre-Training for Esophagitis Detection in Upper GI Endoscopy Images

Tobias Friedetzki, Naveen Chandraiah, Emil Svoboda, Pavel Pecina, Frank Puppe, Adrian Krenzer

Keywords: Esophagitis, Deep Learning, DINO, Upper GI Endoscopy, Self-Supervised

Abstract: Early and accurate detection of esophagitis in upper gastrointestinal endoscopy is essential for guiding targeted treatment and preventing progression to severe diseases such as esophageal cancer. Although deep learning methods have shown promise in supporting esophagitis diagnosis, their performance heavily relies on large amounts of labeled data, which are scarce. Consequently, supervised models often struggle to generalize to the high visual variability and subtle lesion differences encountered in real-world endoscopic examinations. In this work, we study discriminative self-supervised pre-training as a means of leveraging large-scale unlabeled data for robust representation learning. Multiple Vision Transformer models are pre-trained using the DINO framework on 395,201 unlabeled gastrointestinal endoscopy images and subsequently fine-tuned on a curated esophagitis dataset from three clinical centers. Our results demonstrate that self-supervised pre-training on in-domain endoscopic images significantly improves esophagitis detection performance compared to supervised pre-training on natural image datasets such as ImageNet. Specifically, in-domain DINO pre-training yields an average performance gain of 6.60 percentage points in AUPRC on the downstream detection task, with the best-performing model achieving an AUPRC of 89.82%. These findings highlight the importance of in-domain self-supervised learning for reducing annotation dependency and improving model robustness in upper GI endoscopy analysis.

Primary Subject Area: Application: Endoscopy

Secondary Subject Area: Detection and Diagnosis

Registration Requirement: Yes

Reproducibility: https://github.com/tofriede/SSL-Esophagitis-Detection

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 92