EP491 - AI-Assisted Colonoscopy: Where We Are, What the Data Says, and How It Affects the Colorectal Surgeon

Purpose/Background: The development of Artificial Intelligence Colonoscopy (AIC) has allowed the endoscopist a computer assistant for detecting (CADe) and diagnosing (CADx) mucosal irregularities. This technology, while a fairly recent development, has begun rolling out in hospital centers across the globe with the intent to improve precancerous polyp detection rates and reduce rates of malignant degeneration as up to one fourth of colorectal neoplasia are missed on screening colonoscopy. For the colorectal surgeon, the question remains: will we actually improve pathologic progression or are we just removing more benign lesions and increasing cost?

Methods/Interventions: A review of meta-analyses, randomized controlled trials (RCT’s), and practice guidelines was performed using PubMed and Google Scholar, and these data were analyzed for comparison of Polyp Detection Rate (PDR), Adenoma Detection Rate (ADR), Adenoma Miss Rate (AMR), and yield of advanced lesions. Clinical practice guidelines for AI-assisted colonoscopy were also reviewed.

Results/Outcomes: Across several RCT’s the data is clear that AIC can improve PDR and ADR from anywhere between 25–28 % to 32–36 %, and 40–45 % to 50–60 %, respectively, without an increase in statistically significant adverse events. AMR’s were also consistently reduced. Data also suggests an inconsistent yield for detecting flat or serrated lesions, reporting around a 2% increase in these types of lesions. Pooled data also suggests CADe systems resulted in the likelihood of resecting 11% more benign lesions with longer total withdrawal time. The 2025 American Gastroenterological Association guidelines make no recommendation for or against AIC citing that the current data is not certain enough that AIC actually reduces colorectal cancer incidence or mortality.

Conclusion/Discussion: Data supports AIC as a validated endoscopic adjunct when it comes to reducing AMR and improving PDR and ADR across several studies. Improved detection of flat or sessile lesions remains inconsistent, perhaps slightly better. Long term data is lacking when comparing AIC to traditional colonoscopy for colorectal cancer prevention, and because of this, there are no formal guidelines supporting the use of AIC over standard colonoscopy. Surgeons may expect to see stage migration as the software develops, but concerns remain about workflow burden, cost effectiveness, and false positives. The impact of AIC in colorectal surgery will depend largely on further data collection to prove whether AIC reduces cancer rates rather than simply increasing benign resections.