Warning from artificial intelligence against inaccurate polyp size estimation

Effect of an online educational module incorporating real-time feedback on accuracy of polyp sizing in trainees: a randomized controlled trial

“…Kwak et al reported the first use of AI for polyp sizing in 2022 and found that their AI technology improves polyp size estimation when compared with visual estimation (control) [34]. However, some authors have expressed concerns about the limitations of the technology, as sizing using its bifurcation-to-bifurcation distance measuring method varies depending on bowel preparation quality, amount of air insufflation, and presence of peristalsis [35]. The difficulty of AI technologies in predicting size likely lies in the inherent challenge of accounting for depth and size of an in vivo three-dimensional object from a two-dimensional still image on a screen.…”

Measurement of polyp size at colonoscopy: Addressing human and technology bias

“…Assessment of polyp size using the pathologic specimen is not recommended due to disintegration, distortion, and potential shrinkage of the specimen during resection, retrieval, and histologic processing. Yet colonoscopic visual estimation of polyp size is subject to measurement bias and, without a calibrated visual reference, is wildly inaccurate 4–6 …”

“…Technology bias occurs due to image distortion from the colonoscope's fish‐eye lens, where objects in the center of the display appear magnified and objects at the periphery appear small and warped 7 . Makeshift measurement tools, such as forceps or snare catheters, 6 potentially worsen measurement error from technology bias coupled with illusions of relative size, particularly if viewing distance is not considered.…”

“…Yet colonoscopic visual estimation of polyp size is subject to measurement bias and, without a calibrated visual reference, is wildly inaccurate. [4][5][6] Measurement inaccuracy derives from technology and human bias. Technology bias occurs due to image distortion from the colonoscope's fish-eye lens, where objects in the center of the display appear magnified and objects at the periphery appear small and warped.…”

“…A recent application of artificial intelligence (AI) is a computer‐aided measurement tool that is reported 9 and editorialized 6 in a recent issue of Digestive Endoscopy . This technology uses the distance between blood vessel bifurcations in background mucosa as a reference distance (“bifurcation to bifurcation technique”) and is the first published AI‐based measurement technology.…”