Real-Time AI During Pancreatoscopy for Detection of Pancreatic Neoplasia

This prospective pilot study will evaluate the diagnostic performance of a previously validated artificial intelligence (AI) model when applied in real time during digital per-oral pancreatoscopy (POPS). The study will include adults undergoing clinically indicated pancreatoscopy for suspected or known intraductal papillary mucinous neoplasm (IPMN), indeterminate pancreatic-duct abnormalities, or preoperative assessment of IPMN extent. During the procedure, the endoscopist will first record a visual assessment while the AI system is hidden. The AI overlay will then be activated during the same pancreatoscopy examination, and its findings will be recorded independently. AI and endoscopist assessments will be compared with a prespecified reference standard based on surgical histopathology when available or tissue sampling and clinical/imaging follow-up when surgery is not performed. The primary objective is to estimate the sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy of real-time AI for identifying high-grade dysplasia or invasive carcinoma. The study is designed as a pilot to assess feasibility and generate preliminary diagnostic-accuracy estimates for future confirmatory research.