AI in Capsule Endoscopy

Capsule endoscopy lets a patient swallow a small camera that records images as it moves through the gastrointestinal tract, often to investigate possible small-bowel bleeding or other conditions. A complete video may contain many frames and can take a clinician substantial time to review. AI research explores several tasks: finding candidate bleeding or vascular lesions, classifying visual features, measuring how much of the mucosa is visible, and selecting frames that may need closer review. These are aids to interpretation, not autonomous diagnoses. In a 2024 multicenter prospective study of suspected small-bowel bleeding, 133 cases were analyzed after standard and AI-assisted readings. The AI-assisted approach was non-inferior for identifying potentially relevant lesions and had a shorter mean reading time in that study. A separate 90-case study evaluated filtering poorly visualized frames and reported diagnostic agreement with standard reading alongside reduced time. These findings are specific to the systems, indications, readers and study designs tested. They do not establish that every capsule platform can safely skip most frames. A missed lesion can matter, and algorithms may overlook a finding in a poorly prepared segment, confuse artifacts with disease or perform differently across devices. A clinician should review highlighted images, understand the software’s intended use and follow local standards for complete interpretation. AI can make a long video easier to navigate; clinical judgment still determines whether the exam is adequate, what a finding means and whether further testing is needed.