Clinical Report: Suboptimal Performance of Endoscopic Deep Learning Techniques

Overview

This study evaluates the performance of endoscopic image-based classification models for Helicobacter pylori status in gastric cancer patients. The best-performing model achieved a median test AUC of 0.6255.

Background

Helicobacter pylori infection is a significant risk factor for gastric cancer. Traditional diagnostic approaches can be invasive and time-consuming.

Data Highlights

Key Findings

• The study included 602 endoscopic images from 337 gastric cancer patients.
• 576 images from 329 patients were retained for model development and evaluation.
• The best-performing model achieved a median test AUC of 0.6255.
• Model sensitivity was 0.7308, while specificity was 0.3714.
• Fine-tuning the model resulted in a higher median AUC compared to retraining.

Clinical Implications

The limited ability of endoscopic image-based classification to accurately assess H. pylori status in gastric cancer patients suggests a need for further research.

Conclusion

Endoscopic image-based assessment of H. pylori status in gastric cancer patients presents significant challenges, necessitating further research and validation.

Related Resources & Content

1. Surgical Endoscopy, 2024 -- Comparative Efficacy of Various Resolution Endoscopy Techniques for Identifying Mucosal Lesions in the Upper Gastrointestinal Tract
2. Surgical Endoscopy, 2025 -- AI Support Reduces Disparities in Endoscopic Reporting of Gastric Lesions: Results from a Prospective Clinical Study
3. Surgical Endoscopy, 2021 -- Application of Deep Learning Techniques to Endoscopic Images for Evaluating Rectal Cancer Response Following Chemoradiation Therapy
4. Journal of Gastroenterology -- Diagnostic performance of magnifying endoscopy with third-generation narrow-band imaging for early gastric cancer: post hoc analysis of a randomized trial (3G detection trial)
5. ACG H. pylori Guidelines Highlights 2024
6. ESGE Guidelines 2025
7. NCCN Gastric Cancer Guidelines, Version 2.2025
8. AI for Diagnosing H. pylori from Medical Images: A Systematic Review
9. Novel Endoscopic Techniques for Diagnosing H. pylori Infection: A Systematic Review
10. Consensus statements on the current landscape of artificial intelligence applications in endoscopy, addressing roadblocks, and advancing artificial intelligence in gastroenterology - PubMed
11. World Endoscopy Organization Position Statements for Artificial Intelligence in Endoscopic Diagnosis of Gastric Epithelial Neoplasia - PMC
12. Curriculum for safe and effective use of artificial intelligence in endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement - UCL Discovery
13. The Fifth Mexican consensus on the diagnosis and treatment of Helicobacter pylori infection - ScienceDirect
14. Helicobacter pylori prevalence and its spontaneous eradication rate after distal or proximal gastrectomy for gastric cancer: A multicenter prospective cohort study - Omori - 2025 - Annals of Gastroenterological Surgery - Wiley Online Library