An artificial intelligence (AI)-assisted second forward-view examination of the right colon increased the right-colon adenoma detection rate (ADR) by 8.5 percentage points compared with standard colonoscopy in a multicenter randomized controlled trial that included 606 patients in the final analysis, according to a study published in The American Journal of Gastroenterology.
The open-label, prospective trial, conducted at three centers in Japan between July 2023 and October 2025, enrolled patients aged 35 to 75 years undergoing colonoscopy for indications including a positive fecal immunochemical test, surveillance after previous colorectal polypectomy, gastrointestinal symptoms, or opportunistic screening. Participants were randomly assigned 1:1 to an AI-assisted second examination of the right colon or standard colonoscopy, with randomization stratified by study center, age, sex, colonoscopy indication, and endoscopist experience.
In both groups, endoscopists initially examined the right colon from the cecum to the hepatic flexure without AI assistance. They were blinded to group assignment until completing this first examination. In the AI group, the colonoscope was then reinserted into the cecum and a second forward-view examination was performed using the Wise Vision real-time computer-aided detection system. Patients in the standard-colonoscopy group instead continued standard withdrawal from the hepatic flexure to the rectum without a second right-colon examination. The primary outcome was the right-colon ADR, defined as the proportion of patients with at least one adenoma detected in the right colon.
Of 620 patients randomized, 14 were excluded after randomization because of protocol deviations or postrandomization ineligibility, leaving 305 patients in the AI group and 301 in the standard-colonoscopy group. The median age was 60 years, and 43% of patients were female. Baseline characteristics were well balanced between groups.
Primary outcome
Right-colon ADR was 36.4% with AI-assisted reinspection vs. 27.9% with standard colonoscopy, a statistically significant absolute increase of 8.5 percentage points.
“The increase was particularly evident for lesions that can be difficult to detect during routine examination, including flat and sessile lesions and those located in the ascending colon,” lead study author Yusuke Nishikawa, MD, PhD, of the Department of Gastroenterology at Toho University Omori Medical Center in Tokyo, told GI & Hepatology News.
AI-assisted reinspection was associated with 1.48 times the odds of detecting at least one right-colon adenoma. Overall ADR was 60% vs. 52%, respectively, but the difference was not statistically significant.
The right-colon polyp detection rate was also significantly higher with AI-assisted reinspection, at 46% vs. 35%, as was the number of right-colon adenomas per colonoscopy. However, advanced neoplasia detection did not differ significantly between groups (6.6% vs. 7.3%), nor did sessile serrated lesion detection (7.5% vs. 5.3%).
Differences by lesion characteristics
Flat lesions were detected in 24% of patients in the AI group vs. 16% in the standard-colonoscopy group, corresponding to 1.67 times the odds of detection. Sessile lesions were detected in 31% vs. 23%, corresponding to 1.52 times the odds. No significant difference was observed for pedunculated lesions.
For lesions smaller than 5 mm, detection was 29% with AI-assisted reinspection vs. 22% with standard colonoscopy, although this difference did not reach statistical significance. Detection of lesions measuring 5 to 9 mm or at least 10 mm also did not differ significantly.
The strategy increased lesion detection in the ascending colon, where detection rates were 40% with AI-assisted reinspection vs. 29% with standard colonoscopy, representing 1.64 times the odds of detection. No significant difference was found in the cecum.
Endoscopist experience, procedure time, and safety
Exploratory subgroup analyses indicated a similar direction of effect regardless of endoscopist experience. Among experts who had performed at least 1,500 previous colonoscopies, right-colon ADR was 37% with AI-assisted reinspection vs. 29% without it. Among trainees with fewer than 1,500 previous procedures, ADR was 36% vs. 27%, respectively. Neither subgroup difference was statistically significant, and there was no significant interaction between AI use and endoscopist experience.
The additional examination increased examination time. The second AI-assisted right-colon examination had a median duration of 2.5 minutes, and median total right-colon examination time was 5.7 minutes in the AI group vs. 2.9 minutes in the standard-colonoscopy group. Median total colon examination time was also longer, at 13.3 vs. 11.7 minutes, although total procedure time did not differ significantly between groups (median 20.3 vs. 19.7 minutes). No procedure-related or AI-related adverse events were reported.
“These findings suggest that combining AI assistance with a second examination of the right colon may provide a practical strategy for improving the detection of lesions that might otherwise be missed,” said Dr. Nishikawa.
Clinical implications
Dr. Nishikawa further noted that the right colon remains a challenging area for lesion detection because of its complex anatomy and the subtle appearance of many right-sided lesions. “Our findings suggest that combining a simple second forward-view examination with real-time AI assistance could be incorporated into routine colonoscopy practice to improve lesion detection,” he said. “Similar trends were observed across endoscopist experience levels, suggesting that this approach may be applicable in a broad range of routine clinical settings.”
The investigators wrote that the incremental benefit appeared to be mainly related to detection of nonadvanced lesions.
For patients, Dr. Nishikawa said the potential benefit is more complete detection of right-sided adenomas that might otherwise be missed during routine examination. “More reliable detection and removal of these precancerous lesions could potentially contribute to colorectal cancer prevention,” he said. “However, our study did not assess interval colorectal cancer or other long-term patient outcomes, so whether the improved detection translates into a reduction in colorectal cancer will require further investigation.”
Study limitations
The investigators noted that the trial could not determine the independent contribution of AI because the intervention combined AI assistance with a second examination, whereas the standard-colonoscopy group underwent only a single forward-view examination. Baseline ADRs for individual endoscopists were not prospectively collected, and the study evaluated only one AI platform. In addition, the investigators did not assess interval colorectal cancer or other long-term clinical outcomes, and the study focused specifically on the right colon rather than overall colonoscopy performance.
Dr. Nishikawa also noted that future randomized trials directly comparing a conventional second examination with an AI-assisted second examination are needed to determine the independent incremental benefit of AI.
The authors reported no conflicts of interest. The study was supported by a research grant from the Japan Health Promotion Foundation, which had no role in study design, data collection, data analysis, data interpretation, or manuscript preparation.