Research on AI Use in Colonoscopies Produces Mixed Results

Published on: 

Results from two recent studies highlight the potential benefits and limitations of using artificial intelligence technologies in colonoscopies.

A pair of studies examining the potential benefits of artificial intelligence computer-aided detection (CADe) in colonoscopies reached different conclusions regarding the viability of artificial intelligence in these procedures.

As the age of artificial intelligence descends upon medicine, results of the studies, which were a randomized trial and a systematic review and meta-analysis, detail the benefits and shortcomings of current artificial intelligence capabilities in colonoscopies.

Limitations of AI in Colonoscopies

A team of investigators led by Rodrigo Jover, MD, PhD, of Hospital General Universitario de Alicante, conducted a parallel, controlled randomized trial involving 3213 patients from 6 centers in Spain. Named the CADILLAC trial, the study enrolled patients with a positive fecal immunochemical test to assess the contribution of computer-aided detection to colonoscopic detection of advanced colorectal neoplasias, adenomas, serrated polyps, and nonpolypoid and right-sided lesions. Participants were part of a Spanish colorectal cancer screening program and were randomly assigned to a colonoscopy with or without computer-aided detection during the withdrawal phase of the procedure.1

Investigators found no significant difference with the intervention group relative to the control group for detection of advanced colorectal neoplasia detection rate (34.8% vs 34.6%; adjusted risk ratio, 1.01; 95% Confidence Interval [CI], 0.92-1.10), mean number of advanced colorectal neoplasias detected per colonoscopy (0.54 [Standard deviation [SD], 0.95] vs 0.52 [SD, 0.95]; adjusted rate ratio, 1.04; 99.9% CI, 0.88-1.22), or adenoma detection rate (64.2% vs 62.0%; adjusted risk ratio, 1.06; 99.9%, 0.91-1.23).1

Of note, although there was not a significant increase in the detection of advanced lesions, computer-aided detection was associated with a slight increase in the mean number of nonpolypoid lesions (0.56 [SD, 1.25] vs. 0.47 [SD, 1.18] for controls; adjusted rate ratio, 1.19; 99.9% CI, 1.01 to 1.41) and proximal adenomas (0.94 [SD, 1.62] vs. 0.81 [SD, 1.52] for controls; adjusted rate ratio, 1.17; 99.9% CI, 1.03 to 1.33) detected per colonoscopy. Investigators observed similar results in lesions ≤5 mm including polyps (1.68 [SD, 2.42] vs. 1.40 [SD, 2.25] for controls; adjusted rate ratio, 1.20 [99.9% CI, 1.09 to 1.32]), adenomas (1.12 [SD, 1.84] vs. 0.97 [SD, 1.75] for controls; adjusted rate ratio, 1.16 [99.9% CI, 1.04 to 1.30]), and serrated lesions (0.25 [SD, 0.84] vs. 0.19 [SD, 0.68] for controls; adjusted rate ratio, 1.31 [99.9% CI, 1.02 to 1.68]).1

“The current findings are a snapshot of what these systems can currently offer and what can be expected from them. Detecting more advanced lesions still lies in the hands of experienced endoscopists who can recognize the lesions and achieve adequate mucosal exposure,” wrote Jover et al.1

In Support of AI in Colonoscopies

A group of investigators from Humanitas University in Milan, Italy, led by Alessandro Repici, MD, director of the Gastroenterology and Digestive Endoscopy Unit in Humanitas Research Hospital, conducted a systematic review and meta-analysis of 21 randomized trials comparing CADe with standard colonoscopy for polyp and cancer detection. Trials were extracted from Medline, Embase, and Scopus databases by 2 reviewers and further analyzed for benefit and harm outcomes.2

“Before widespread implementation of AI tools using CADe, systemic evaluation and quantification of both benefits and potential harms is warranted,” wrote Repici et al.2

For the purpose of analysis, investigators identified adenoma detection rate (ADR), number of adenomas detected per colonoscopy, number of serrated lesions per colonoscopy, adenoma miss rate, and advanced adenoma (APC) as benefit outcomes. Investigators defined advanced adenoma as an adenoma of 10 mm or greater with high-grade dysplasia and villous histology. Number of polypectomies for nonneoplastic lesions and withdrawal time were extracted as harm outcomes. Studies were pooled using a random-effects model for each outcome and certainty of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation framework.2

The use of CADe for polyp detection during colonoscopy resulted in increased detection of adenomas compared to the standard colonoscopy group (44.0% vs. 35.9%; relative risk, 1.24 [95% CI, 1.16 to 1.33]), corresponding to a 55% (risk ratio, 0.45 [CI, 0.35 to 0.58]) relative reduction in miss rate. Investigators pointed out this increased detection was limited to diminutive (≤5 mm) adenomas and was not observed in cases of advanced adenomas.2

Investigators also noted higher rates of unnecessary removal of nonneoplastic polyps in the CADe group than the standard group (0.52 vs. 0.34 per colonoscopy; mean difference [MD], 0.18 polypectomy [CI, 0.11 to 0.26 polypectomy]). Mean inspection time was also slightly increased with CADe compared to the control groups (9.22 vs. 8.73 minutes).2

In an editorial, Dennis Shung, MD, MHS, PhD, assistant professor of medicine at Yale University, expressed optimism regarding the potential of artificial intelligence in colonoscopies, but noted current gaps related to understanding and real-world performance limit the immediate applicability of the technology.3

“The current gap between randomized controlled trial performance and real-world performance is concerning but likely reflects both differences in clinician behavior outside of trials and the complexity of real-world clinical environments,” Shung wrote.3 “More broadly, the gap reveals the socio- technical challenge of algorithmic integration into heterogeneous and often messy clinical workflows. How algorithmic systems partner with clinicians and how these should be designed and refined across heterogeneous systems and contexts are necessary questions that must be explored to minimize disruption and lead to real-world effectiveness.”


1. Mangas-Sanjuan C, de-Castro L, Cubiella J, et al. Role of Artificial Intelligence in Colonoscopy Detection of Advanced Neoplasias : A Randomized Trial [published online ahead of print, 2023 Aug 29]. Ann Intern Med. 2023;10.7326/M22-2619. doi:10.7326/M22-2619

2. Hassan C, Spadaccini M, Mori Y, et al. Real-Time Computer-Aided Detection of Colorectal Neoplasia During Colonoscopy : A Systematic Review and Meta-analysis [published online ahead of print, 2023 Aug 29]. Ann Intern Med. 2023;10.7326/M22-3678. doi:10.7326/M22-3678

3. Shung DL. From Tool to Team Member: A Second Set of Eyes for Polyp Detection [published online ahead of print, 2023 Aug 29]. Ann Intern Med. 2023;10.7326/M23-2022. doi:10.7326/M23-2022