Artificial Intelligence Versus Conventional Scoring for Assessing Small Bowel Capsule Endoscopy Cleanliness in...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Assessing small bowel cleanliness in capsule endoscopy is crucial but challenging.
• The KODA score is validated but time-consuming for routine use.
• An AI tool (AXAROlite) was developed to automate this assessment.

Purpose of the Study:

• To evaluate the performance of the AI-based AXAROlite tool against the KODA score for small bowel cleanliness in Crohn's disease (CD) patients.
• To compare the accuracy, correlation, and agreement between AXAROlite and the KODA score.
• To investigate clinical factors affecting AI-assessed cleanliness.

Main Methods:

• Post hoc analysis of a multicenter randomized controlled trial (NCT05117996).
• Included 142 CD patients undergoing capsule endoscopy after standard or simplified bowel preparation.
• Videos were assessed using the KODA score and the AI tool AXAROlite.

Main Results:

• AXAROlite and KODA scores showed strong correlation (ρ=0.61, p<0.001).
• AXAROlite demonstrated high accuracy (AUROC=0.85) in detecting adequate cleanliness (KODA > 2.25).
• Lower AI scores were associated with active disease and longer transit time.

Conclusions:

• AXAROlite offers a rapid, automated, and accurate method for evaluating small bowel cleanliness in CD patients.
• The AI tool's performance is comparable to the established KODA score.
• Integrating AXAROlite could streamline reporting and reduce variability in clinical practice.