Classification of Biliary Strictures Using Real-Time Cholangioscopy Artificial Intelligence: The SMART-AI Trial

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Biliary stricture classification is crucial for patient management but often inaccurate with current sampling techniques.
• Previous studies showed artificial intelligence (AI) analyzing static cholangioscopy images outperformed sampling methods.
• A need exists for real-time AI tools to improve diagnostic accuracy during procedures.

Purpose of the Study:

• To prospectively evaluate a real-time cholangioscopy AI system in a single-center trial.
• To compare the AI's biliary stricture classification performance against standard sampling techniques (brush cytology, forceps biopsy).
• To assess the AI's accuracy relative to human endoscopists (junior and experienced).

Main Methods:

• A prospective trial involving 41 patients with suspected biliary strictures.
• A cholangioscopy AI system analyzed video streams in real-time during procedures.
• Primary outcome: AI vs. sampling techniques accuracy. Secondary outcome: AI vs. 14 human observers accuracy.

Main Results:

• The cholangioscopy AI achieved 87.8% accuracy in classifying biliary strictures.
• AI accuracy (87.8%) was significantly higher than sampling techniques (67.4%; p=0.043).
• AI outperformed junior (61.5%; p=0.001) and experienced (63.1%; p=0.011) endoscopists.

Conclusions:

• Traditional sampling techniques and human interpretation of cholangioscopy have limitations in biliary stricture classification.
• Real-time cholangioscopy AI demonstrates superior accuracy compared to current methods.
• AI systems may offer significant benefits for aiding biliary stricture diagnosis.