SUNColonoscopyVid - sporedata/researchdesigneR GitHub Wiki

General description

Japan's SUN (Shizuoka University of Nursing and Medical Care) Colonoscopy Video Database is a collection of colonoscopy videos aimed at advancing the development of computer-aided diagnosis (CAD) systems and improving the overall quality of colonoscopic procedures. The database contains a large volume of high-quality colonoscopy footage, capturing various types of colorectal polyps, lesions, and normal colon mucosa.

The SUN Colonoscopy Video Database provides a substantial number of videos with diverse characteristics, which helps in training and testing AI models for recognizing colorectal polyps and cancerous lesions. It is thus a valuable resource for advancing research in AI-assisted endoscopy, enhancing diagnostic accuracy, and providing training for healthcare professionals, contributing significantly to early detection and prevention of colorectal cancer.

Limitations

  1. Data Privacy: Maintaining the privacy of patients is paramount. Videos in the database are anonymized to protect patient identities, adhering to strict ethical guidelines.
  2. Complexity and Variability: The database includes videos with a range of complexities, such as different levels of bowel preparation, lighting variations, and the presence of artifacts (e.g., stool, bubbles). This helps train models to perform well in varied real-life conditions, but it also presents challenges to model robustness and generalizability.
  3. Real-Time Assistance: CAD systems developed using the SUN database have applications in providing real-time feedback to endoscopists during procedures, helping reduce fatigue-related errors.

Related publications

Data access

For more information on the SUNColonoscopy dataset, visit http://amed8k.sundatabase.org/