AI-assisted system improves biliopancreatic disease detection

A novel AI-assisted endoscopic ultrasound (EUS) system demonstrated superior performance in recognizing biliopancreatic stations and segmenting anatomical structures. Testing with 45,737 EUS images revealed the mean teacher algorithm achieved 95.60% accuracy in internal tests and 92.72% in external validation for station recognition. Additionally, U-Net v2 emerged as the optimal model for segmentation. Notably, the system outperformed mid-level endoscopists, potentially expediting learning curves and standardizing EUS procedures.