CAD-CAP: a 25,000-image database serving the development of artificial intelligence for capsule endoscopy.

Area of Science:

• Medical Imaging
• Gastroenterology
• Computer Science

Background:

• Capsule endoscopy (CE) is the primary method for small bowel (SB) examination.
• Reading CE videos is time-consuming due to a large number of frames per video (mean 50,000).
• Development of computer-assisted diagnosis (CAD) tools is needed to improve efficiency.

Purpose of the Study:

• To describe the creation and structure of the CAD-CAP database.
• To facilitate the development of CAD tools for capsule endoscopy reading.
• To address the challenges of analyzing large volumes of CE data.

Main Methods:

• A multicenter retrospective study involving 12 French endoscopy centers.
• Inclusion of third-generation SB-CE videos (Pillcam, Medtronic).
• Manual segmentation and classification of pathological frames by expert readers and automated extraction of normal frames.

Main Results:

• The CAD-CAP database contains 4,174 SB-CE videos.
• 1,480 videos (35%) had pathological findings, with 5,184 pathological frames extracted.
• Extracted frames included fresh blood, vascular lesions, and inflammatory/ulcerative lesions; 20,000 normal frames were also included.
• The database has supported the development of automated angiectasia detection tools and international challenges.

Conclusions:

• The CAD-CAP database is a valuable resource for developing CAD tools in capsule endoscopy.
• It aids in the efficient analysis of small bowel capsule endoscopy videos.
• Facilitates advancements in automated medical image analysis for gastroenterology.