Adenoma Detection Rate in Artificial Intelligence-assisted Colonoscopy

Colorectal cancer (CRC) is the third most common cancer, and the second most common cause of cancer-related death worldwide. CRC screening is used for detection and removal of precancerous lesions before they develop into cancer. Colonoscopy is regarded being superior to other screening tests, and is therefore used as the golden standard. Screening colonoscopy is associated with a reduced risk of CRC-related death. Since it is not possible for an endoscopist to determine the histopathology of the polyp with certainty during a colonoscopy, detected pre-malignant lesions should be removed and sent for histological examination. Multiple studies have shown that there is a strong association between findings at the baseline screening colonoscopy and rate of serious lesions at the follow up colonoscopy. Risk factors for adenoma, advanced adenoma and cancer at follow-up colonoscopy are multiplicity, size, villousness, and high degree dysplasia of the adenomas at the baseline screening colonoscopy. The adenoma detection rate (ADR) is the percentage of examinations performed by one endoscopist, in which one or more adenomas are found. This is widely accepted as the main quality indicator for each endoscopist and colonoscopy. There is strong evidence that the ADR is inversely correlated to the incidence of interval CRC. With each 1,0% increase in the ADR there is a 3,0% decrease in the risk of developing CRC. Unfortunately, adenomas and advanced adenomas are frequently missed, and the ADR varies widely among different endoscopists. Also, the quality changes throughout the day. Both the withdrawal time and the ADR decreases by the end of the day, approximately by 20% and 7% respectively. Small improvements in the colonoscopy quality may have great importance for the outcome when screening for CRC. Artificial intelligence (AI) can reduce the performance variability by working as a pair of additional virtual eyes, compensating for perceptual errors due to fatigue, distraction and inaccurate human vision. Within the last few years there have been published several randomized controlled trials (RCT) investigating the efficacy of real time computer-aided detection. Among these, all of the RCT´s which have ADR as the primary outcome, have shown that the use of AI contributes to a significantly higher ADR, compared colonoscopies without assistance of an AI system. Repici et al. have shown that experience of the endoscopist only plays a minor role as a determining factor. Correspondingly, results from a previous study by Liu et al. indicates that CADe systems are not only useful for endoscopists with a low detection rate, but can also increase the ADR for more experienced endoscopists. Kamba et. al reports a significant lower adenoma miss rate (AMR) for CADe-assisted colonoscopy, compared to a conventional colonoscopy. This is independent on the endoscopist´s level of expertise. Other studies conclude that AI probably will benefit the less experienced endoscopists more. However, there are only a limited number of studies investigating the impact of AI when used by less experienced endoscopists. According to a recent RCT from Wallace et al. the use of AI can reduce the AMR by approximately 50%, but primarily due to increased detection of small (\<10 mm) flat neoplasia. This difference is slightly higher than in a previous study, in which the relative reduction was approximately 35%. However, in this study there were no significant difference in missed diminutive polyps (\<10 mm). In a systematic review the overall withdrawal time was shown to be higher with AI-assisted colonoscopy (AIC), compared to conventional colonoscopy (CC), but the ADR and PDR was also higher. Naturally, there have been concerns about prolonged colonoscopy time, and increased workload if implementing the AI system, since the increased detection of small polyps may lead to unnecessary polypectomy. However, two recent RCT´s report that the unnecessary resection of non-neoplastic polyps did not increase by using the CADe system. The results so far are promising, suggesting that AIC is superior to CC when it comes to polyp and adenoma detection. Routine use of computer-aided polyp detection (CADe) systems could further reduce the incidence of interval CRC, but more clinical data from large multicenter randomized trials are required to understand the actual impact of AI in the daily clinical setting. We have designed a quality assurance multicenter quasi RCT to investigate the effect of real time AI-assistance (GI Genius, Medtronic) on adenoma detection rate (ADR) in both experienced and less experienced endoscopists. We want to investigate whether the CADe system can reduce the performance variability and increase the ADR significantly. The overall aim of this research is to investigate if AI-assistance in colonoscopy can increase the ADR. This prospective, multicenter, quasi randomized controlled trial (RCT) will take place at four endoscopy units in Region Zealand, Denmark. These units are located at Zealand University Hospital (Køge), Nykøbing Falster Hospital, Holbæk Hospital and Næstved Hospital. All units except Næstved Hospital are participating in the national CRC-screening programme. We will screen all patients scheduled for screening, diagnostic, and surveillance colonoscopy. The eligible patients will receive a colonoscopy from an expert or a non-expert endoscopist based on the normal distribution of endoscopists at the endoscopic units.