A Pilot Study on Automatic Three-Dimensional Quantification of Barrett's Esophagus for Risk Stratification and...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Barrett's esophagus diagnosis relies on Prague C&M classification, which is subjective and operator-dependent.
• Accurate measurement of Barrett's Epithelium Area (BEA) and associated risk scores is crucial for patient management.
• Existing methods lack automated quantification and 3D visualization capabilities.

Purpose of the Study:

• To develop and validate a novel artificial intelligence (AI) system for automated measurement of Barrett's esophagus.
• To enable accurate quantification of BEA and islands, and 3D reconstruction of the esophageal surface.
• To assess the system's accuracy on both phantom and real patient endoscopic data.

Main Methods:

• A deep learning-based depth estimator network predicts endoscope camera distance.
• Segmentation of BEA and gastroesophageal junction, projected to estimated distances for C&M score calculation.
• Validation using a 3D printed esophagus phantom and 194 patient endoscopic videos.

Main Results:

• Phantom data showed 97.2% accuracy for C&M/island measurements (±0.9 mm deviation) and 98.4% accuracy for BEA (±0.4 cm² deviation).
• Patient data analysis revealed AI-generated C&M scores concurred with expert endoscopists, with low mean differences (8% for C, 7% for M).
• The system achieved high accuracy in automated Prague C&M score extraction.

Conclusions:

• The proposed AI methodology automatically extracts Prague C&M scores with high accuracy.
• Automated quantification and 3D reconstruction of Barrett's area offer novel avenues for risk stratification.
• This technology has the potential to enhance the assessment of therapy response in Barrett's esophagus patients.