AI-Driven Colon Cleansing Evaluation in Capsule Endoscopy: A Deep Learning Approach

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Minimally invasive diagnostic methods are advancing in gastroenterology.
• Colon capsule endoscopy (CCE) is a key first-line diagnostic tool for colon evaluation.
• Effective bowel preparation is critical for CCE, but current classification scales have limited interobserver agreement.

Purpose of the Study:

• To develop and validate a deep learning algorithm for automated classification of colonic bowel preparation for CCE.
• To improve the objectivity and reliability of bowel preparation assessment in CCE.

Main Methods:

• Development of a deep learning algorithm (neural network) for automatic bowel preparation classification.
• Evaluation of the algorithm's performance using sensitivity, specificity, accuracy, and area under the curve (AUC).

Main Results:

• The deep learning algorithm achieved high performance: 91% sensitivity, 97% specificity, and 95% overall accuracy.
• The algorithm demonstrated strong discriminating capacity with AUC values between 0.92 and 0.97.
• The developed classification system is easily applicable.

Conclusions:

• Automated classification of bowel preparation using deep learning is feasible and highly accurate.
• This AI-driven approach is essential for the widespread adoption of CCE.
• The algorithm supports the integration of minimally invasive panendoscopy in clinical practice.