Beyond intuition: Establishing an expert-derived reference standard for real-time AI guidance in ERCP

Deep selective biliary cannulation is the fundamental yet most technically challenging aspect of endoscopic retrograde cholangiopancreatography (ERCP) [1], with failure rates of up to 20% [2]. Repetitive attempts increase post-ERCP adverse events. While artificial intelligence (AI) has transformed general endoscopy [3], ERCP AI applications remain largely prognostic or restricted to anatomy classification (e.g., difficulty scoring, predicting adverse events) [4–9]. Currently, no AI provides real-time, intraoperative cannulation guidance.