Background: With the rapid advance of artificial intelligence (AI) technology, AI approved as a medical device is being introduced into daily clinical practice. Several studies reported the efficacy of Computer-Aided Detection (CADe) in colonoscopy for adenoma detection. However, these studies were mainly conducted at academic centers. In this study, we evaluated the efficacy of CADe in colonoscopy in community-based practices.
Methods: We conducted a single-center, open-label randomized controlled trial in a Japanese community hospital. Patients aged ≥40 years who underwent colonoscopy for screening, post-polypectomy surveillance, fecal immunohistochemical test positivity, or symptoms such as abdominal pain or hematochezia were randomly assigned in a 1:1 ratio to standard colonoscopy or CADe (EndoBRAIN-EYE) assisted colonoscopy. Randomization was stratified by gender, age, and experience of endoscopists. Patients with inadequate bowel preparation (0 or 1 in any 1 of the 3 segments in the Boston Bowel Preparation scale) or inability of the endoscope to reach the cecum were excluded from the analysis. The primary outcome was adenoma detection rate (ADR). Secondary outcomes were adenomas per colonoscopy (APC), sessile serrated lesion detection rate (SDR), sessile serrated lesions per colonoscopy (SPC), and withdrawal time.
Results: 1043 patients were recruited between September 2022 and August 2023. After excluding 29 patients who did not meet inclusion criteria or declined to participate, 505 patients were randomly assigned to standard colonoscopy, and 509 patients to CADe assisted colonoscopy. Furthermore, patients with inadequate bowel preparation or inability of the endoscope to reach the cecum were excluded. Finally, 497 patients in the standard group and 501 patients in the CADe group were analyzed. Baseline characteristics between the 2 groups were similar. There was no significant difference in ADR (standard vs CADe: 54.5% vs 50.7%, P = 0.278), while APC were significantly higher in the standard group (1.34 vs 1.14, P = 0.007). SDR (9.5% vs 9.8%, P = 0.986) and SPC (0.14 vs 0.13, P = 0.903) were similar in the standard and CADe groups. The mean withdrawal time was significantly longer in the CADe group (691 seconds vs 751 seconds, P = 0.020).
Conclusions: In community-based practices, CADe assisted colonoscopy did not improve the detection of colorectal neoplasia. Moreover, CADe assisted colonoscopy resulted in lower adenomas per colonoscopy. Further research is needed to better understand why some endoscopists benefit from CADe and others do not. (UMIN Clinical Trials Registry number, UMIN000049054)