Introduction

Artificial intelligence systems, also known as computer-aided detection (CADe) systems, have been developed to improve lesion detection rate during colonoscopy. After initial reports of high efficacy, there is an increasing recognition of variability in the effectiveness across different CADe systems. Moreover, recent real-world studies of commercially available systems reported comparable adenoma detection rates between CADe and conventional colonoscopy (CC). The aim of this study was to evaluate a CADe system, the DISCOVERY^TM system (PENTAX Medical, Tokyo, Japan), in a diagnostic, non-iFOBT screening, and surveillance colonoscopy population.

Aims & Methods

A multicenter, randomized, controlled trial (RCT) was conducted at 7 hospitals (both university and non-university) in Europe and Canada. Patients referred for diagnostic, non-iFOBT screening, or surveillance colonoscopy were included. Patients were randomized (1:1) to undergo CADe-assisted colonoscopy or CC by experienced endoscopists (>500 independently performed colonoscopies). Patients with a total Boston Bowel Preparation Scale (BBPS) <6 were excluded from the final analysis. Primary outcome was adenoma detection rate (ADR). Secondary outcomes included mean adenoma per colonoscopy (APC), polyp detection rate (PDR) and mean sessile serrated lesion per colonoscopy (SSLPC).

Results

In total, 567 patients were enrolled, of which 497 were included in the final analysis; 250 in the CADe arm and 247 in the CC arm. Surveillance was the indication in 202/497 (40.6%) colonoscopies, followed by diagnostic in 199/497 (40.0%), and non-iFOBT screening in 96/497 (19.3%), with equal distribution between both study arms. Overall, ADR was similar between CADe vs. CC; 38.4% (96/250) vs. 37.7% (93/247), p=0.86. APC was also similar between CADe vs. CC; 0.66 (166/250) vs. 0.66 (163/247), p=0.97. CADe increased PDR, albeit not statistically significantly; 55.2% (138/250) vs. 51.4% (127/247), p=0.398. Notably, SSLPC was increased in the CADe group vs. CC; 0.30 (76/250) vs. 0.19 (46/247), p=0.049.

Conclusion

In our study conducted by experienced endoscopists, CADe did not result in a statistically significant increase in adenoma detection (ADR or APC) in patients with variable indications for colonoscopy. Remarkably, the detection of sessile serrated lesions was significantly increased by 58%, suggesting that this CADe system can detect these more subtle lesions which are increasingly recognized for their clinical importance, even when used by experts.

Introduction

Computer-aided polyp detection (CADe) systems based on artificial intelligence are increasingly used in routine endoscopic colorectal cancer screening. These systems are regularly updated and occasionally provide adjustable detection thresholds that impact their performance. However, little is known about the effect of these adjustments on the system’s performance.

Aims & Methods

The objective of this study is to evaluate the performance of various CADe systems on a common benchmark dataset, and to compare the performance of different versions and configurations of the same system. The study employed a total of 101 colonoscopy videos, and each video containing visible polyps was manually annotated with a bounding box, resulting in a total of 129 705 polyp images. The performance of three different CADe systems was assessed: two versions of GI Genius, two detection types of EndoAID, and the freely available CADe system EndoMind. The evaluation included a comprehensive analysis of sensitivity, false positive rate, and time to first polyp detection. In addition, the quality of the bounding boxes in delimiting the polyp was also evaluated by calculating the intersection over the union.

Results

EndoAID (type A), the earlier version of GI Genius, and EndoMind detected all 93 polyps in at least one image. Both the later version of GI Genius and EndoAID (type B) missed one polyp. The mean per-frame sensitivity for each system was 50.63 % for the earlier version of GI Genius, 67.85 % for the later version of GI Genius, 65.60 % for EndoAID (type A), 52.95 % for EndoAID (type B), and 60. 22 % for EndoMind. Regarding the mean first detection time, the earlier version of GI Genius required 1 510 ms, the later version of GI Genius required 607 ms, EndoAID (type A) required 659 ms, EndoAID (type B) required 1 316 ms, and EndoMind required 1 083 ms. The CADe system that presented the fewest number of false positive frames was EndoAID (Type B) with a ratio of only 0.63 %. Lastly, EndoMind was the best performing system in terms of bounding box quality with an intersection over union value of 68.32 %.

Conclusion

This study provides a direct comparison of the performance of different CADe systems, as well as different versions and configurations of the same system. The results show that the performance of CADe systems can vary significantly depending on the system used, the version of the system, and the configuration of the system. By providing an objective evaluation of the strengths and weaknesses of CADe systems, this study can help clinicians and researchers select the most appropriate system for their specific needs.

Introduction

Computer-aided detection (CADe) systems have gained attention for their use in colonoscopy, with growing evidence of increased adenoma detection rates (ADR). However, results remain inconsistent, and debate regarding their usefulness continues, including their impact on serrated lesion (SL) detection and performance across varying levels of endoscopist expertise. Given that CADe is used as a real-time assistive tool, it is crucial to confirm its efficacy in lesion detection before considering long-term outcomes.

Aims & Methods

We conducted a multinational, multicenter randomized controlled trial in the Asia-Pacific (Project CAD/NCCH2217; jRCT1032230396) to assess the efficacy of CADe in improving lesion detection during colonoscopy performed as part of colorectal cancer (CRC) screening [1]. We enrolled asymptomatic individuals aged 50–79 years undergoing either primary screening colonoscopy or colonoscopy following a positive fecal immunochemical test (FIT), who had not received colonoscopy in the past 5 years. Participants were randomly assigned (1:1) to the CADe group (colonoscopy with CADe assistance) or the standard group (colonoscopy without CADe) using a minimization method to balance age, sex, colonoscopy indication, endoscopist experience, and country/region. In the CADe group, approved CADe devices were used in accordance with each country's regulations, including EndoBRAIN-EYE, ENDO-AID, CAD EYE, and WISE VISION. All procedures were performed by endoscopists with experience of ≥500 colonoscopies. The primary endpoint was adenoma detection rate (ADR). Secondary endpoints included adenomas per colonoscopy (APC), advanced ADR (AADR), sessile serrated lesion detection rate (SSLDR), sessile serrated lesions per colonoscopy (SSLPC), scope withdrawal time, and the resection rate of non-neoplastic lesions. Planned sample size was 1400 cases (one-sided α = 0.05, 80% power).

Results

Of the 1,399 participants enrolled between December 28, 2023, and March 6, 2025, 1,373 were included in the full analysis set (688 and 685 in the CADe and standard groups, respectively). In both groups, the proportions of primary screening and post-FIT colonoscopies were 75% and 25%, respectively. No differences were observed between the CADe and standard groups in ADR (54.7% vs. 53.9%; odds ratio [OR] 1.04; 95% confidence interval [CI], 0.83–1.30, one-sided p=0.38), APC (1.17 vs. 1.24; risk ratio [RR], 0.94; 95% CI, 0.81–1.09), or AADR (5.5% vs. 7.6%; OR, 0.71; 95% CI, 0.46–1.10). The CADe group showed higher SSLDR (9.0% vs. 6.0%; OR, 1.55; 95% CI, 1.03–2.33) and SSLPC (0.11 vs. 0.07; RR, 1.65; 95% CI, 1.09–2.50). Subgroup analysis based on endoscopist experience demonstrated a higher SSLDR in the CADe group among those with 500–4,999 (10.1% vs. 5.4%) and 5,000–9,999 (11.9% vs. 3.6%) prior colonoscopies, but not among those with experience of ≥10,000 procedures (7.0% vs. 7.5%). There was no difference between groups in the non-neoplastic lesion resection rate (14.7% vs. 13.4%; OR, 1.11; 95% CI, 0.82–1.50), and the median scope withdrawal time in cases without polypectomy was 9 minutes in both groups.

Conclusion

CADe assistance did not improve adenoma detection where high ADR was already achieved with standard colonoscopy. Nevertheless, it enhanced SSL detection, except among highly experienced endoscopists, without prolonging withdrawal time or increasing unnecessary polypectomies. While our findings do not overstate the efficacy of CADe, they support a reasonable and practical role for its use in improving colonoscopy quality in CRC screening.

References

1. Sekiguchi M, et al. Protocol for a multicenter randomized controlled trial to assess the usefulness of computer-aided detection systems for colonoscopy in colorectal cancer screening in the Asia-Pacific region (project CAD/NCCH2217). Jpn J Clin Oncol. 2025; 55: 837-40.

Disclosure

The authors have no conflict of interest related with this study.