Artificial Intelligence-assisted Colonoscopy in the Detection and Characterization of Colorectal Lesions

Overview

The study aims to evaluate the effectiveness of artificial intelligence-assisted colonoscopy in increasing adenoma detection rate and the accuracy in the characterization of colorectal lesions, compared to standard colonoscopy, in a randomized controlled clinical trial setting.

Colorectal cancer (CRC) currently shows, according to GLOBOCAN, an incidence of 19.5 individuals per 100,000 inhabitants in both sexes, being the third most common cancer in men and the second in women, representing the third leading cause of death in both men and women. According to the GLOBOCAN registry of the World Health Organization (WHO), it is estimated that CRC is the third most common type of cancer worldwide, responsible for 10% of all newly diagnosed cancer cases, corresponding to 1,931,590 cases in 2020, preceded only by lung cancer (11.4%) and breast cancer (11.7%). CRC is the second leading cause of cancer mortality (9.4%; 935,173 cases in 2020), following only lung cancer, which accounts for 18% of cancer deaths globally. In Brazil, according to data from the National Cancer Institute (INCA), CRC mirrors the global incidence, being the second most common cancer by sex. Colonoscopy is the most accurate CRC screening method, with sensitivity reaching 100% in the detection of colorectal lesions. According to studies, for each 1% increase in adenoma detection rate, there is a 5% decrease in CRC mortality, highlighting the importance of performing colonoscopy to detect colorectal lesions, especially adenomas. Consequently, with the advancement of technology, new high-definition endoscopes with virtual chromoscopy and image magnification have been developed to increase adenoma detection rates. More recently, AI-assisted colonoscopy has been gaining prominence in helping prevent CRC in some medical centers worldwide, such as in Japan. In a multicenter study with 700 patients in 2019, a significantly higher adenoma detection rate was demonstrated with AI-assisted colonoscopy compared to standard colonoscopy (54.8% vs. 40.4%). Subsequently, a randomized, double-blind clinical trial with 1,058 patients was conducted, comparing standard colonoscopy to AI-assisted colonoscopy. The result was an adenoma detection rate of 29% for AI-assisted colonoscopy and 20% for standard colonoscopy, with the difference being statistically significant. Two other studies comparing AI-assisted colonoscopy and standard colonoscopy showed similar results. However, when analyzing the accuracy of AI systems in characterizing colorectal lesions, different results are observed in the literature. On one hand, Japanese studies report accuracies above 90% in characterizing neoplastic and non-neoplastic lesions with artificial intelligence, while other studies, such as the Dutch study and the German study, found accuracies of 74.4% and 84.7%, respectively, results significantly lower compared to the Japanese studies. Therefore, given not only the differences in results obtained by various authors but also the differences in population and the lack of studies on AI-assisted colonoscopy in developing countries, the objective of this work is to evaluate the adenoma detection rate of AI-assisted colonoscopy and assess the accuracy of artificial intelligence in characterizing colorectal lesions.