AI-Assisted Detection and Reporting of Colonic Angiodysplasia

Colonic angiodysplasia is a common finding during colonoscopy, yet it is frequently left undocumented. Despite being one of the leading causes of lower gastrointestinal bleeding, it is often noted informally or omitted from reports altogether, creating gaps in patient care and follow-up.

To address this, we developed an AI system that tackles the problem from two angles. During the procedure itself, a computer vision model analyses the colonoscopy feed in real time, flagging angiodysplasia as it appears on screen. After the examination, a locally deployed large language model cross-checks the written report to verify whether the finding has been properly documented. Together, the two components form a closed loop: one ensuring findings are seen, the other ensuring they are recorded.