Computer‐aided detection in colonoscopic practice: Ready for prime time?

Abstract

Because the majority of colorectal cancers can be prevented by endoscopic polypectomy due to the slow-growing nature of the precursors, 1 postcolonoscopy colorectal cancer or interval cancer should receive attention. Coloscopy is the global standard to screen the large intestine but a nonnegligible portion of colorectal polyps, including sessile serrated lesions, have been missed during colonoscopy. According to the latest systematic review and meta-analysis, 2 amazingly, 26% of adenoma and 9% of advanced adenoma were missed in recent reports. Reasons for this misidentification can be attributed to inadequate distension of the colon, suboptimal cleansing, insufficient inspection time, and other factors. Even when these obstacles are removed; however, colonoscopists occasionally misrecognize the lesions despite their endoscopic visibility because colonoscopy with high concentration of polyps to seek sessile serrated lesions and flat lesions is a challenging task at any time. Although the use of chromoendoscopy, narrow band imaging, blue laser imaging, or linked color imaging is one of the solutions for this misidentification, artificial intelligence can also be a strong solution because of its ease of application.CAD-EYE (Fujifilm, Tokyo, Japan) is one of the computer-aided diagnosis (CAD) systems commercially available in Japan and Europe. In Japan, more than three CAD systems provided by Olympus, NEC, and LPIXEL are commercially available, although the CAD system provided by Medtronic has not obtained regulatory approval from the Japanese government. To my knowledge, in Japan, there is only one randomized controlled trial (RCT) using the CAD system developed by LPIXEL and Jikei University to date. 3 As for the CAD-EYE system, two RCTs 4,5 have been published in the past 2 years, both successfully demonstrating a significantly higher adenoma detection rate (ADR) as the primary outcome. In contrast, Yamaguchi et al. 6 conducted an RCT to confirm the performance of CAD-EYE but failed to demonstrate a significant difference in ADR, only in adenoma miss rate (AMR) of novice endoscopists.So far, there have been numerous RCTs investigating the performance of CAD systems published in the past