Artificial Intelligence for Colorectal Polyp Detection and Characterization

AbstractPurpose of reviewTo elucidate the advantages and limitations of existing artificial intelligence technologies for colonoscopy by evaluating the currently available eight prospective studies (two for automated polyp detection and six for automated polyp characterization).Recent findingsAI is expected to mitigate the inherent risk of human error causing a polyp to be missed or mischaracterized by assisting polyp detection and characterization (i.e., pathological prediction). Some of the prospective studies clearly demonstrate the potential for AI to improve adenoma detection rates, which is considered one of the most important quality indicators for colonoscopies, or achieve a>  90% negative predictive value in differentiating diminutive (≤ 5 mm) rectosigmoid adenomas which is considered as a threshold required for optical diagnosis. However, it is also important to consider the negative impacts of AI, such as the deskilling effect on healthcare providers, which has yet to be sufficiently addressed.SummaryWe believe that AI can become standard practice in colonoscopy procedures within several years, given its rapid spread and its expected low implementation cost. However, considering the limited evidence supporting the use of AI for colonoscopy, additional studies should be done to explore the long-term efficacy and safety of AI in colonoscopy and implement robust endpoints such as colorectal cancer incidence and mortality.
Source: Current Treatment Options in Gastroenterology - Category: Gastroenterology Source Type: research