The potential of deep learning for gastrointestinal endoscopy—a disruptive new technology

Abstract Applications of deep learning have the potential to revolutionize gastrointestinal endoscopy. When trained by experts and capable of real-time feedback, deep learning can be applied to improve disease detection; to assist interventions; and to document procedure findings, interventions, and quality measures. In colonoscopy, deep learning is already showing promise for polyp detection, polyp characterization, documentation of complete exams, calculating withdrawal times, quantifying preparation quality, and identifying tools used for intervention. Similarly, in video capsule endoscopy, deep learning is showing great potential to reduce miss rates, time-to-find, and reading times. The technical challenges of live implementation are rapidly disappearing with inexpensive high-performance graphics-processing units. We can soon expect to enjoy an “expert in the room” helping all endoscopists perform at high levels, as well as an “expert personal scribe” allowing clinicians to replace documentation time with more face time with patients.