Detection of elusive polyps using a large-scale artificial intelligence system (with videos)

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Colorectal cancer is a major cause of mortality.
• Colonoscopy is the gold standard for detecting and removing precancerous polyps.
• The polyp miss rate in colonoscopies ranges from 22% to 28%.

Purpose of the Study:

• To evaluate the performance of the DEEP DEtection of Elusive Polyps (DEEP2) system.
• To assess the system's effectiveness in detecting elusive polyps during colonoscopies.

Main Methods:

• The DEEP2 system was trained on 3611 hours of colonoscopy videos and validated on 1393 hours.
• Ground truth was established by expert gastroenterologists reviewing videos offline.
• A prospective clinical validation study involved 100 procedures.

Main Results:

• DEEP2 achieved 97.1% sensitivity for all polyps with 4.6 false alarms per video.
• The system detected an average of 0.22 additional polyps per sequence not identified by human endoscopists.
• In clinical practice, DEEP2 identified an average of 0.89 extra polyps per procedure.

Conclusions:

• DEEP2 demonstrates high sensitivity in polyp detection.
• The system effectively increases polyp detection rates in both video analysis and real-time colonoscopies.
• DEEP2 offers a promising tool for improving colonoscopy outcomes with minimal false alarms.