Effect of Real-time Computer-aided System (ENDO-AID) on Adenoma Detection in Endoscopist-in-training: a Single-blind Randomized Study

Colorectal cancer (CRC) is the most common and second most lethal cancer in Hong Kong with more than 5,600 new cases and 2,300 deaths annually. Colonoscopy with polypectomy has shown to reduce CRC-related mortality by 53%. However, high polyp miss rates were reported to be up to 26% for adenomas and 9% for advanced adenomas in standard colonoscopies. Risk factors included proximal location, serrate or flat lesions, poor bowel preparation and short withdrawal time (<6 minutes). Insufficient trainee experience was also associated with a higher adenoma miss rate. A significant proportion of interval CRC was attributed to the missed lesions during index colonoscopy leading to adverse patient outcomes.

As a result, various techniques were developed to improve adenoma detection rate (ADR) during colonoscopies. Techniques including water exchange method, second examination of the right colon (retroflexion or second forward view)and cap/cuff-assisted colonoscopies were proven to increase ADR effectively. However, these techniques were operator-dependent requiring certain level of expertise.

Recently, artificial intelligence and computer-aided polyp detection (CADe) systems have developed rapidly around the globe. These systems can provide real-time CADe by flagging the suspected lesions to endoscopists, with the adoption of deep learning or convoluted neural networks. A number of prospective randomized clinical trials reported a significant increase in ADR in CADe group. The number of adenoma detected per colonoscopy was consistently higher among different polyp sizes, location and morphology. The ADR increment was particularly higher for diminutive adenomas smaller than 5mm.

Nevertheless, most of the aforementioned studies only involved senior endoscopists for the procedures. Theoretically, the senior endoscopists were more skillful to expose colonic mucosa and more experienced to distinguish the false positive computer signals, leading to an enhanced performance of CADe in real-time colonoscopies. The effect of CADe on inexperienced junior endoscopists performing colonoscopies remains largely unknown.

In this single-blind randomized study, the investigators aim to evaluate the effect of a new CADe system (ENDO-AID) on adenoma detection and quality improvement in junior endoscopists.