Cloud-based AI system enhances detection of adenomas, ‘difficult-to-detect’ polyps
Click Here to Manage Email Alerts
Key takeaways:
- The adenoma detection rate was 8% higher with Caddie, a cloud-based AI system, vs. standard of care.
- Proximal ADR also increased from 15% to 23% with Caddie.
A cloud-based artificial intelligence system significantly improved adenoma detection, including hard-to-detect lesions in the proximal colon, by non-expert endoscopists in the United Kingdom, according to results presented at UEG week.
“When you look at previous randomized controlled trials in this field, the majority are performed by expert and experienced endoscopists,” Rawen Kader, MRCP, PhD, a clinical research fellow at University College London, said during his presentation. “However, the main benefit in AI polyp detection was always hoped to be in our non-expert cohort endoscopists to try and level the playing field.”
Kader and colleagues aimed to determine whether Caddie (Odin Vision), a real-time, cloud- based AI polyp detection could improve ADR when performed by non-expert endoscopists in routine clinical settings.
From April 2021 to December 2022, researchers enrolled 739 adults who were scheduled for surveillance or symptomatic colonoscopy, of whom 614 were included in analysis and assigned to standard of care (n = 309) or intervention with Caddie (n = 306).
The procedures were performed by 26 endoscopists at nine academic and community hospitals in the U.K., each of whom was required to have performed a minimum of 100 colonoscopies with a baseline ADR of less than 30%. To reduce study bias, Kader noted each endoscopist was limited to a maximum of 60 procedures.
Study end points included the difference in endoscopist ADR and detection of adenomas per colonoscopy between procedure arms, as well as polyps per colonoscopy and neoplastic detection rate.
“For the primary outcome, there was an 8% increase of the adenoma detection rate for the non-expert endoscopists when using the Caddie system,” Kader said. “This increased from 25% to 33%.”
The proximal ADR also significantly increased with the Caddie system compared with standard of care from 15% to 23%, Kader said. Further, the neoplastic detection rate increased 9% with Caddie system from 27% to 36%.
On a per polyp basis, the odds of detecting adenomas was 50% higher with the Caddie system vs. standard of care (0.66 vs. 0.42).
“When you look at all types of polyps, you see that the increase was 33%,” Kader said. “What is really interesting in these findings is that the significant increase of detection of polyps with the Caddie system is in the more difficult-to-detect polyps: the flat non-polypoid sessile polyps and the polyps we tend to find in the proximal colon.”
Adenomas per extraction (67.5% vs. 60.6%) and false positive rate (10.7% vs. 11.7%) were not significantly different between the Caddie and standard of care arms, Kader noted, “although crucially, you can see that numerically both these results favored the Caddie system.”
“Each incremental increase in ADR is thought to correlate with a reduced risk of a patient developing a post-colonoscopy colorectal cancer, or PCCRC,” Kader told Healio. “The risk of PCCRC is higher in non-expert endoscopists, and AI may prove critical in reducing the inter-operator variability in ADR and the prevalence of PCCRC.”
Perspective
Back to Top
Christine Lee, MD
Colorectal cancer is the second leading cause of cancer death in the United States, and colonoscopy remains the gold standard in colon cancer prevention. We continually endeavor to improve the quality of colonoscopies, including integrating innovations in technology.
Kader and colleagues aimed to determine whether a cloud-based, artificial intelligence polyp-detection system could improve adenoma detection rate when performed by nonexpert endoscopists in routine clinical settings in the United Kingdom.
The increase in ADR is statistically significant but how this ultimately benefits patients is unclear, given that the diminutive adenomas more likely to be detected by the AI system are of relatively lower risk for metachronous advanced neoplasia. Whether this translates into more lives saved needs to be further examined.
However, it is very compelling that the “neoplastic detection rate increased by 9%” with the AI system (35-33%). This warrants additional studies for reproducibility and whether it can be extrapolated to the U.S.
Collapse
Kader R, et al. Randomized controlled trial of a cloud-based artificial intelligence polyp detection system (Caddie). Presented at: UEG Week; Oct. 14-17, 2023; Copenhagen, Denmark (hybrid meeting).
Read more about
Click Here to Manage Email Alerts