Artificial intelligence in small bowel capsule endoscopy - current status, challenges and future promise.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in gastroenterology

Background:

• Physicians face time-consuming small-bowel capsule endoscopy reviews.
• Computer-assisted detection aims to improve efficiency and accuracy.

Purpose of the Study:

• To explore the development of neural network-based solutions for capsule endoscopy.
• To assess the potential of automated characterization and physician acceptance of AI tools.

Main Methods:

• Review of current neural network-based solutions for computer-assisted detection.
• Discussion of the need for large labeled databases and expert ground truth definitions.
• Exploration of AI developments for landmark localization, lesion measurement, and bowel cleanliness assessment.

Main Results:

• Weakly supervised solutions show promise but lack video-level evaluations and prospective studies.
• Automated characterization requires supervised machine learning and extensive labeled data.
• Ongoing developments focus on anatomical landmark identification and lesion analysis.

Conclusions:

• AI holds potential to streamline capsule endoscopy reviews and improve diagnostic accuracy.
• Further validation through prospective studies and large datasets is crucial.
• The market integration and physician/patient acceptance of AI tools remain key considerations.