Artificial intelligence for diagnosing gastric lesions under white-light endoscopy

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Endoscopist diagnostic accuracy for gastric lesions varies, with potential for missed or misdiagnosed cases.
• Workload, fatigue, and subtle lesion appearances under white-light endoscopy (WLE) contribute to diagnostic challenges.

Purpose of the Study:

• To develop an artificial intelligence (AI) system for diagnosing six common gastric lesions using WLE images.
• To evaluate the AI system's performance and its role in assisting endoscopists.

Main Methods:

• A dataset of 29,809 WLE images from 8947 patients was used for training, and 1579 images from 496 patients for testing.
• The AI system's diagnostic performance was compared against 12 endoscopists, with and without AI assistance.

Main Results:

• The AI system achieved an overall accuracy of 85.7%, comparable to senior endoscopists (85.1%) and superior to junior endoscopists (78.8%).
• AI assistance significantly improved diagnostic accuracy for both senior (89.3%) and junior (86.2%) endoscopists.
• Diagnostic times were slightly reduced with AI assistance, though not significantly.

Conclusions:

• The developed AI system serves as a valuable auxiliary tool for endoscopists.
• AI assistance can enhance diagnostic accuracy and potentially reduce endoscopist workload in identifying gastric lesions.