VIENNA ― With computer-aided detection (CADe), more adenomas were found, and ther was a twofold reduction in missed adenomas for patients undergoing screening or surveillance colonoscopy, according to results from a multicenter, randomized, controlled trial that compared it with conventional colonoscopy.
The researchers found that with the addition of CADe, the adenomas per colonoscopy (APC) rate increased from 0.51 to 0.70, which was statistically significant. The adenoma detection rate (ADR) increased from 30% to 37%, and the adenoma miss rate (AMR) decreased from 36% to 19%.
The study is unique in its testing of CADe in daily clinical care, and the results support use in this setting.
The findings were presented by gastroenterologist and study lead Michiel Maas, MD, from the Department of Gastroenterology and Hepatology, Radboudumc, Nijmegen, the Netherlands, during a late-breaker session at the United European Gastroenterology (UEG) Week 2022 meeting.
The reason for the increased detection rates was that operators who used CADe found more 0- to 5-mm adenomas, as well as more adenomas in the proximal colon, compared to those who performed conventional colonoscopy, Maas reported.
"This is in line with other studies," he said.
Maas also reported that CADe performed better in finding 6- to 9-mm adenomas, which are more clinically relevant. Adenomas larger than that were not very prevalent in the study population, he noted.
Fewer Misses With AI Tool
Although conventional colonoscopy is considered the gold standard for detection and removal of adenomas, around 26% are missed. CADe shows promise in raising the ADR, explained Maas.
Their study aimed to evaluate a novel CADe system in patients referred for screening (non–immunochemical-based fecal occult blood test) or surveillance colonoscopy (3–10 years since their index colonoscopy).
The study was conducted across 10 sites in the United States, Europe, and Israel. Patients were randomly assigned to undergo either conventional colonoscopy (467 patients) or CADe (449 patients).
Patients were matched, and the mean age of the patients was 60 years. In the conventional colonoscopy group, 45% were women; 44% underwent colonoscopy as surveillance, and 56% underwent colonoscopy for screening. In the CADe group, 48% were women; 43% underwent colonoscopy for surveillance; and 57% underwent colonoscopy for screening.
The primary outcome was adenoma per colonoscopy. The rate was 0.7 among people who underwent CADe, compared to 0.51 among those screened with conventional colonoscopy. This difference reached statistical significance (P = .014).
The ADR, which was a secondary outcome, was higher with CADe than with conventional colonoscopy (37% vs 30%; P = .014), while the AMR was lower, at 19% vs 36%.
In addition, a subset of patients (15%) underwent tandem colonoscopies, either conventional colonoscopy followed by CADe (69 patients) or vice versa (64 patients). In this tandem design, to better calculate the adenoma miss rate, the same practitioner performed both colonoscopies. The AMR was 19% in the CADe-first group, compared to 36% in the conventional colonoscopy-first group (P = .024).
Maas offered reasons as to why CADe increased adenoma detection.
"In the computer-aided colonoscopy, the practitioner still reads the image in the conventional way, but the AI algorithm also reads the image and displays in real time a box around the suspected polyp to aid the practitioner in detecting more adenomas," Maas said.
"The AI is a second pair of eyes, but it pays the same attention to every pixel, whereas a human endoscopist has to focus on something, and the subtle, smaller lesions might be missed easily," he said.
With human endoscopy, operator error can be a factor, Maas added.
"To an extent, AI helps to level the playing field between operators, although AI still relies on the images that are visualized and provided for analysis," he said.
Maas noted that the value of finding diminutive adenomas is still uncertain. "Long-term research is needed to determine whether it is worth detecting and acting on these findings," he said.
"We also need more long-term research to evaluate the effect of AI on colorectal cancer. We are hopeful that AI will lead to better detection during colonoscopy and thus improve the overall quality," he said.
Mixed Results for Advanced Lesions
The session also featured a presentation on a multicenter, randomized, controlled trial called the CADILLAC study, which evaluated the role of AI for detection of advanced lesions during colonoscopies.
As opposed to detecting small polyps, the value of which is controversial, the CADILLAC study investigated whether CADe was useful in the detection of advanced and more significant lesions, explained presenter Carolina Mangas-Sanjuan, MD, a gastroenterologist from the Hospital General Universitario de Alicante, Spain.
"The aim is different in this trial," said Mangas-Sanjuan. She noted that the investigators questioned whether it is desirable to detect small lesions that are probably not important or big lesions that are more relevant to the incidents of colorectal cancer.
The study involved patients who were part of the Spanish population–based colorectal cancer screening program and who had been identified as being at high risk for advanced lesions via positive results on a fecal immunochemical test. The participants were randomly assigned to undergo colonoscopy either with assistance of CADe or without it.
The 3213 patients (1610 in the CADe group and 1603 in the control group) were similar in their baseline characteristics. The mean age of the participants was 61 years; 53.4% were men, and the mean Boston Bowel Preparation Scale (BBPS) score was 7.8.
The primary outcome was detection of an advanced colonic lesion (advanced adenoma and/or advanced serrated polyp).
There were no differences in the detection of advanced colonic lesions (CADe group, 34.8%, vs control group, 34.6%; adjusted odds ratio [aOR], 1.01; 95% CI: 0.87 – 1.16).
The mean number of advanced colonic lesions per colonoscopy was not increased through use of CADe (CADe group, 0.54 ± 0.95, vs control group, 0.52 ± 0.95; 95% CI: -0.05 to 0.09), Mangas-Sanjuan reported.
"This might be because advanced colonic lesions are normally easy to see," she suggested.
Despite the high detection rates in the control group, there were differences in some of the secondary outcomes, for example, the detection of large (≥10 mm) serrated lesions (CADe group, 5.3%, vs 3.8% with conventional colonoscopy; aOR, 1.44; 95% CI: 1.03 – 2.03), nonpolypoid lesions (27.8% vs 24.1%; aOR, 1.21; 95% CI: 1.03 – 1.42), and proximal adenomas (43.7% vs 38.2%; aOR, 1.26; 95% CI: 1.09 – 1.46).
Use of AI to aid colonoscopy is "in a dynamic phase, and there is a need for improvement in this technology, using larger and more variant datasets to train deep-learning systems," Mangas-Sanjuan concluded.
Questions Still Remain
After the session, the moderators shared their thoughts on the application of CADe in the identification of adenomas during colonoscopy.
"Computer-aided detection is definitely here to stay," said Barbara A. J. Bastiaansen, MD, from Amsterdam UMC, the Netherlands.
Helmut Messmann, MD, University Hospital Augsburg, Augsburg, Germany, added that gastroenterologists are learning how to implement CADe into their daily practice. "The discussion is becoming about when should we use CADe and should we be training fellows in it," he said.
"We need to understand more about whether AI can help us reduce costs, and also in the future, how we will use AI to assist in treatment decisions, for example, whether we need to resect polyps. But identifying lesions is the first step," Messmann added.
The value of identifying diminutive lesions is an important topic of debate, said Bastiaansen.
"Asking whether we really want to find these very small polyps and what is the effect on long-term colorectal cancer risk and mortality are common criticisms. However, it is possible that if we remove many small polyps in one colonoscopy, then it may pay off in the long run because we can lengthen the surveillance intervals," she said.
Both Messman and Bastiaansen agreed that long-term studies are needed to answer these questions.
United European Gastroenterology (UEG) Week 2022: Abstracts LB06 and LB09. Presented October 10, 2022.
Magentiq Eye LTD (Israel) was sponsor of the study presented in abstract LB06. The study presented in abstract LB09 was supported by a research grant from Medtronic. Maas, Mangas-Sanjuan, Bastiaansen, and Messmann report no relevant financial relationships.
For more news, follow Medscape on Facebook, Twitter, Instagram, and YouTube.
Lead image: Medscape Illustration: Getty Images
Medscape Medical News © 2022
Cite this: AI Tool Plus Colonoscopy Catches More Adenomas in Clinical Care - Medscape - Oct 12, 2022.