Computer aided polyp detection has limited clinical efficacy

Abstract

Role of artificial intelligence in identifying colorectal cancer is still evolving Colorectal cancer (CRC) is the third most common cancer worldwide, often arising from precancerous adenomas.12 Therefore, adenoma detection rate (ADR) is a key quality measure for colonoscopy, with higher ADRs associated with an improved survival benefit.34 Given the importance of this metric, advancements in artificial intelligence (AI) have led to the development of computer aided polyp detection (CADe) systems aimed at improving ADR. The BMJ Rapid Recommendations panel reviewed 44 trials on CADe for polyp detection, highlighting a pooled 8% increase in ADR compared with standard colonoscopy but noting no direct evidence on patient-important outcomes such as CRC incidence or mortality.5 An accompanying microsimulation model by Halvorsen et al concluded that CADe significantly increased surveillance recommendations after screening colonoscopy (by 6.37%) and modestly increased recommendations for colonoscopy after a positive faecal immunochemical test (FIT) (by 0.82%).6 The model predicted that implementation of CADe colonoscopy would prevent one additional CRC per 1000 individuals undergoing screening colonoscopy and five CRCs per 10 000 individuals with a positive FIT screening test followed by colonoscopy over 10 years. However, these studies underscore the limited evidence for a clinically important decrease in …