AI-Assisted Evaluation of Colon Cleanliness in Capsule Endoscopy Videos

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate colon capsule endoscopy (CCE) video evaluation is crucial for diagnosing gastrointestinal disorders.
• Assessing the cleanliness of the colon is a key step in determining the admissibility of CCE videos for analysis.
• This study addresses the need for improved efficiency and accuracy in colon cleanliness scoring.

Purpose of the Study:

• To introduce and evaluate a novel system for assisting physicians in assessing colon cleanliness scores from CCE videos.
• To improve the objectivity and consistency of colon cleanliness evaluations.

Main Methods:

• A TransUNet architecture was employed for video analysis.
• A customized loss function and a low-effort labeling method were utilized.
• The system was evaluated on a dataset of 52 CCE videos, with agreement measured using Cohen's kappa statistic.

Main Results:

• The system achieved a Cohen's kappa agreement of 0.586 with physicians.
• This agreement surpasses the observed intra-observer variability (0.546).
• The system visualizes cleanliness evolution throughout the video, aiding score justification.

Conclusions:

• The developed system shows potential to support colon cleanliness evaluation in CCE.
• The system demonstrates superior agreement compared to inter-physician agreement.
• Visualizing cleanliness evolution enhances the transparency and interpretability of AI-generated scores.