Masterclass: Advanced GI Endoscopic Imaging (Live/Virtual) | December 2025-Parsa - Artificial Intelligence in Capsule Endoscopy and Small Bowel

Parsa - Artificial Intelligence in Capsule Endoscopy and Small Bowel

Pdf Summary

This presentation by Dr. Nasim Parsa discusses the integration of Artificial Intelligence (AI), particularly deep learning (DL) and machine learning (ML), into capsule endoscopy (CE) for small bowel disease detection and diagnosis. Capsule endoscopy involves reviewing hours of video footage, with only seconds showing relevant pathology, making it challenging for endoscopists due to the time, cost, and interpretation skills required.<br /><br />AI tools assist in identifying small bowel lesions such as bleeding (arteriovenous malformations, AVMs), ulcers, strictures, and neoplasms. Notably, studies have demonstrated that AI models can process CE images to detect Crohn’s disease lesions, intestinal strictures, and differentiate disease pathology with improved efficiency and accuracy. For example, one DL model analyzing nearly 28,000 images from multiple patients showed high accuracy in classifying strictures, ulcers, and normal mucosa.<br /><br />Two main training approaches are highlighted: one training on images from different patients than the test set to ensure generalizability but lower accuracy, and another training on images that include the same patients, which risks bias but improves diagnostic performance. AI explainability is enhanced by class activation maps, which highlight regions in images driving the model’s classifications.<br /><br />A significant benefit is the drastic reduction in reading time for CE videos from about 97 minutes to under 6 minutes, improving productivity by approximately 16-fold while slightly improving diagnostic accuracy. This positions AI as a powerful assistive tool that complements, rather than replaces, human expertise in gastroenterology.<br /><br />Overall, AI implementation in capsule endoscopy promises to address challenges in modern endoscopy such as interpreting complex data and limited expert availability, ultimately enhancing clinical throughput and diagnostic precision in small bowel disease management.

Keywords

Artificial Intelligence

Deep Learning

Machine Learning

Capsule Endoscopy

Small Bowel Disease

Lesion Detection

Crohn's Disease

Diagnostic Accuracy

Class Activation Maps

Reading Time Reduction