Artificial Intelligence in Upper Gastrointestinal Endoscopy: Current Evidence, Practice, and Future Directions.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Upper gastrointestinal endoscopy (UGIE) workflows are evolving with AI integration.
• AI tools like computer-aided detection (CADe) and diagnosis (CADx) aim to enhance quality control and diagnostic accuracy.
• Current AI applications span various conditions including Barrett's neoplasia, esophageal cancer, gastric cancer, and Helicobacter pylori assessment.

Purpose of the Study:

• To review recent advancements in AI for UGIE.
• To discuss AI performance in key upper gastrointestinal applications.
• To critically evaluate challenges and future directions for AI clinical implementation.

Main Methods:

• Review of deep learning models, including convolutional neural networks and reinforcement learning.
• Analysis of AI performance in lesion detection, invasion-depth estimation, and characterization.
• Evaluation of AI's impact on image review acceleration and diagnostic accuracy.

Main Results:

• Deep learning models demonstrate high sensitivity and specificity, potentially matching expert performance in identifying gastrointestinal lesions.
• AI shows promise in accelerating image review and improving targeted sampling over random biopsies.
• AI applications are effective across various conditions like Barrett's neoplasia, esophageal squamous cell carcinoma, early gastric cancer, and H. pylori assessment.

Conclusions:

• AI has the potential to standardize UGIE, boost diagnostic confidence, and improve training without replacing clinical judgment.
• Successful clinical adoption hinges on prospective validation, bias mitigation, workflow integration, and appropriate training.
• Human-in-the-loop oversight, interpretable AI outputs, and cost-effective deployment are crucial for realizing AI's benefits in patient care.