Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Buonocore MR, Cesaro P, Conforti F, Consonni D, Cottone I, Dabizzi E, Mantia B, Monica F, Mutignani M, Olivari N, Picardi G, Salvi D, Scaramella L, Schettino M, Topa M, Vecchi M, Elli L. Impact of computer-aided polyp detection in screening and diagnostic colonoscopy: A multicenter randomized controlled trial. World J Gastroenterol 2026; 32(29): 116893 [DOI: 10.3748/wjg.116893]
Corresponding Author of This Article
Gian Eugenio Tontini, MD, PhD, Assistant Professor, Department of Pathophysiology and Organ Transplantation, University of Milan, Via della Commenda 19, Milan 20122, Lombardy, Italy. gianeugenio.tontini@unimi.it
Research Domain of This Article
Gastroenterology & Hepatology
Article-Type of This Article
research-article
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Baishideng Publishing Group Inc, 7041 Koll Center Parkway, Suite 160, Pleasanton, CA 94566, USA
Share the Article
Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Buonocore MR, Cesaro P, Conforti F, Consonni D, Cottone I, Dabizzi E, Mantia B, Monica F, Mutignani M, Olivari N, Picardi G, Salvi D, Scaramella L, Schettino M, Topa M, Vecchi M, Elli L. Impact of computer-aided polyp detection in screening and diagnostic colonoscopy: A multicenter randomized controlled trial. World J Gastroenterol 2026; 32(29): 116893 [DOI: 10.3748/wjg.116893]
Gian Eugenio Tontini, Giovanni Aldinio, Giulia Picardi, Matilde Topa, Maurizio Vecchi, Luca Elli, Department of Pathophysiology and Organ Transplantation, University of Milan, Milan 20122, Lombardy, Italy
Gian Eugenio Tontini, Francesco Conforti, Emanuele Dabizzi, Lucia Scaramella, Maurizio Vecchi, Luca Elli, Gastroenterology and Endoscopy Unit, Foundation IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milan 20122, Lombardy, Italy
Lorenzo Dioscoridi, Irene Cottone, Massimiliano Mutignani, Digestive and Interventional Endoscopy Unit, ASST Niguarda, Milan 20162, Lombardy, Italy
Raffaele Salerno, Beatrice Mantia, Mario Schettino, Division of Gastroenterology, ASST Fatebenefratelli Sacco, Milan 20162, Lombardy, Italy
Matteo Rossano Buonocore, Fabio Monica, Department of Gastroenterology and Digestive Endoscopy, Academic Hospital Cattinara, Trieste 34149, Friuli Venezia Giulia, Italy
Paola Cesaro, Nicola Olivari, Daniele Salvi, Gastroenterology and Digestive Endoscopy Unit, Poliambulanza Brescia Hospital, Brescia 25124, Lombardy, Italy
Dario Consonni, Occupational Health Unit, Foundation IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milano 20122, Lombardy, Italy
Author contributions: Tontini GE, Dioscoridi L, Salerno R, Vecchi M, and Elli L contributed to conceptualization; Consonni D contributed to data curation and formal analysis; Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Buonocore MR, Cesaro P, Conforti F, Cottone I, Dabizzi E, Mantia B, Monica F, Mutignani M, Olivari N, Salvi D, Scaramella L, Schettino M, Topa M, Vecchi M, and Elli L contributed to investigation; Tontini GE, Consonni D, Vecchi M, and Elli L contributed to methodology; Tontini GE, Vecchi M, and Elli L contributed to project administration and supervision; Tontini GE, Dioscoridi L, Salerno R, Vecchi M, and Elli L contributed to validation; Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Topa M, Vecchi M, and Elli L contributed to visualization; Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Buonocore MR, Cesaro P, Conforti F, Consonni D, Cottone I, Dabizzi E, Mantia B, Monica F, Mutignani M, Olivari N, Picardi G, Salvi D, Scaramella L, Schettino M, Topa M, Vecchi M, and Elli L contributed to writing of original draft, and review and editing. All authors have read and approve the final manuscript.
Supported by Italian Ministry of Education and Research - MUR; Italian Ministry of Health, No. RC2026/150/01.
Institutional review board statement: The research protocol was reviewed and approved by the Institutional Review Board of Foundation IRCCS Ca’ Granda Ospedale Maggiore Policlinico, No. OSMAMI-15/09/2022-0042892-U.
Clinical trial registration statement: The trial was registered on ClinicalTrials.gov, No. NCT07171333.
Informed consent statement: Written informed consent was obtained from all participants prior to their inclusion in the study, in accordance with the ethical standards of the Declaration of Helsinki.
Conflict-of-interest statement: Dr. Tontini reports PENTAX Medical Europe loaned the DISCOVERY™ systems for the duration of the study.
CONSORT 2010 statement: The authors have read the CONSORT 2010 Statement, and the manuscript was prepared and revised according to the CONSORT 2010 Statement.
Data sharing statement: Data collected with deidentified participants will be available from the corresponding author upon request.
Corresponding author: Gian Eugenio Tontini, MD, PhD, Assistant Professor, Department of Pathophysiology and Organ Transplantation, University of Milan, Via della Commenda 19, Milan 20122, Lombardy, Italy. gianeugenio.tontini@unimi.it
Received: November 24, 2025 Revised: February 13, 2026 Accepted: April 22, 2026 Published online: August 7, 2026 Processing time: 235 Days and 21.7 Hours
Abstract
BACKGROUND
Colonoscopy is the reference standard for the diagnosis of polyps and neoplasms of the lower gastrointestinal tract. Recently, real-time artificial intelligence algorithms have been developed to minimize human error in detecting lesions during colonoscopy. Although the performance of these systems on still images and real-time video appears extremely promising, robust evidence of their ability to improve everyday clinical practice is still lacking.
AIM
To evaluate the use of a computer-aided detection (CADe) system on polyp detection rate, adenoma detection rate (ADR), advanced ADR, and sessile serrated lesion detection rate, in addition to ultra-high-definition (UHD) systems in screening and diagnostic procedures, stratified by expertise and fatigue.
METHODS
Between August 2023 and September 2024, patients were randomized (1:1) to UHD colonoscopy with or without CADe (DISCOVERY™) at five Italian tertiary centers. Endoscopists were categorized by annual volume (< 200 or ≥ 200 procedures) and assessed for work-related fatigue (i.e., daily working hours and number of procedures).
RESULTS
Of 421 randomized patients, 409 were included in the per-protocol population. The CADe group were similar in polyp detection rate [CADe: 41.2% vs conventional colonoscopy (CC): 45.4%], ADR (CADe: 29.4% vs CC: 32.2%), advanced ADR (CADe: 5.9% vs CC: 5.4%), or sessile serrated lesion detection rate (CADe: 6.3% vs CC: 8.7%). Results remained consistent across different indications, operator expertise, and fatigue levels.
CONCLUSION
CADe did not improve detection rates over UHD colonoscopy in tertiary centers. No robust evidence of effect modification by operator volume or workload-related fatigue was observed, although the findings in low-volume endoscopists may warrant further investigation in larger studies.
Core Tip: This study shows that in tertiary centers, computer-aided detection (CADe) did not enhance overall polyp or adenoma detection compared with conventional, ultra-high-definition colonoscopy. High-volume endoscopists did not benefit and in some cases performed slightly worse with CADe. Indicators of operator fatigue did not significantly influence CADe effectiveness.
Citation: Tontini GE, Dioscoridi L, Salerno R, Aldinio G, Buonocore MR, Cesaro P, Conforti F, Consonni D, Cottone I, Dabizzi E, Mantia B, Monica F, Mutignani M, Olivari N, Picardi G, Salvi D, Scaramella L, Schettino M, Topa M, Vecchi M, Elli L. Impact of computer-aided polyp detection in screening and diagnostic colonoscopy: A multicenter randomized controlled trial. World J Gastroenterol 2026; 32(29): 116893
Colorectal cancer (CRC) is a leading diagnosis and cause of death worldwide[1]. Early-onset CRC (< 50 years) is growing, raising concerns about modifying screening programs[2]. Colonoscopy is the gold standard for CRC screening and diagnosis, with a negative exam allowing 10-year rescreening[3]. Improving colonoscopy quality indicators is crucial to enhance accuracy and reduce post-colonoscopy CRC[4]. Adenoma detection rate (ADR), a key metric (target ≥ 25%), reflects the percentage of patients with ≥ 1 histologically confirmed adenoma or carcinoma and inversely correlates with post-colonoscopy CRC risk[5,6]. ADR varies with operator experience[7], competence[7-9], and fatigue[10], and declines as endoscopy sessions progress[11,12]. Endoscopic imaging can also affect ADR, by unveiling subtle mucosal lesion with the help of high-definition (HD), ultra-HD (UHD), image-enhanced endoscopy techniques and computer-aided detection (CADe) systems[13-16].
The adoption of CADe systems in the everyday endoscopic practice is expected to increase the overall ADR by implementing high-quality standards and supporting endoscopists in challenging cases. However, published studies report inconsistent findings on CADe effectiveness in ADR improvement. Specifically, several randomized controlled trials (RCTs) and meta-analyses report an increase in ADR with CADe, but the real-life studies seem to show less favorable results[6,17-20]. Remarkably, the available evidence comes from studies that were not designed to measure the impact of CADe beyond the advantages offered by UHD imaging integrated with artificial intelligence (AI) systems.
Furthermore, conflicting data exist regarding variations in effectiveness based on operator expertise: The available evidence suggests that CADe may provide greater support to less experienced endoscopists[17]. Additionally, literature lacks studies focusing on the impact of CADe according to operators’ fatigue, despite its known effect on performance[11,12]. Based on these considerations, we designed this study to elucidate the contentious aspects of AI-assisted colonoscopy in a real-life scenario.
MATERIALS AND METHODS
We conducted a randomized, multicenter trial across five Italian tertiary referral centers. Consecutive patients undergoing a screening or diagnostic colonoscopy were randomized 1:1 to UHD CADe-assisted colonoscopy or UHD conventional colonoscopy (CC). Demographics and indications were recorded, classifying patients into screening (primary, secondary, surveillance) or diagnostic categories (Supplementary material). The study was reported following the Consolidated Standards of Reporting Trials guidelines[21].
Endoscopists were classified as low- (< 200/year) or high-volume (≥ 200/year); fatigue was assessed using indirect workload-related indicators (i.e., working hours, number of procedures). These variables were used as pragmatic proxy measures of fatigue, rather than validated assessments of physical or cognitive fatigue. Patients in the CADe arm underwent colonoscopy with CADe DISCOVERY™ (PENTAX Medical, Tokyo, Japan), highlighting suspected lesions in real time (sensitivity 90%, specificity 80%, CE mark in January 2020[22,23]). The final lesion characterization remained the endoscopist’s responsibility.
Eligible patients were aged 40-80 years without prior colorectal surgical resection, recent diverticulitis, inflammatory bowel disease, familial polyposis, and a negative colonoscopy within five years. In addition, patients with one of the following post-randomization exclusion criteria were excluded from study analyses: Inadequate bowel preparation (i.e., Boston Bowel Preparation Scale < 2 in any segment)[14], inadequately discontinued antithrombotic/antiplatelet therapy, and non-resected or non-retrieved polyps leading to incomplete data collection[24].
All colonoscopies were conducted using PENTAX Medical (Tokyo, Japan) full HD endoscopes (i.e., 1920 × 1080 pixels; e.g., 90i, i10, i10c, i20c, Pentax series), full HD video processors (i.e., 1920 × 1080 pixels; EPK7000, Optivista plus EPK-i7010) or UHD video processors (3840 × 2160 pixels; Inspira EPK-i2080c). All procedures, either in CC or in the CADe group, were displayed on 32-inch UHD monitors (3840 × 2160 pixels) hereby “upscaling” full HD images in a 4k format consistent with the state of the art based on commercially available equipment.
All procedures were performed following a standardized protocol to improve colorectal neoplasia detection (e.g., permanent use of i-Scan 1, ≥ 6-minute withdrawal, position changes) but no mechanical devices for improving detection (e.g., Cap, Endocuff, G-EYE®) were permitted[13]. Patients were randomized by endoscopists, using REDCap, to CADe or CC and stratified post hoc into four subgroups: High-volume endoscopists (HVE) or low-volume endoscopists (LVE), with CADe or CC.
Recorded parameters included bowel preparation, intubation time, withdrawal time, total procedure time, polyp detection rate (PDR), ADR, advanced ADR (AADR), sessile serrated lesion (SSL) detection rate (SSLDR), adenomas per colonoscopy (APC), and SSL per colonoscopy. The primary endpoint was the ADR. Secondary endpoints included the PDR, colonoscopy withdrawal time, and ADR/PDR stratified according to operator fatigue indicators (i.e., working hours, number of procedures)[11,12] and endoscopists’ experience (< or ≥ 200 colonoscopies per year).
Statistical analysis
We planned to enroll 400 patients, with 200 examined using CADe and 200 without CADe. For HVE, the expected ADR was 35% without CADe and 40% with CADe, corresponding to an absolute increase of 5% (five additional patients with adenomas per 100 examined). For LVE, the expected ADR was 20% without CADe and 40% with CADe, corresponding to an absolute increase of 20% (20 additional patients with adenomas per 100 examined). Thus, the expected difference in ADR gain between LVE and HVE was a 15% [95% confidence interval (CI): 6%-24%]. In summary, with this sample size, we anticipated that for every 100 patients, LVE using CADe would have detected adenomas in approximately 15 more patients than HVE (95%CI: 6-24). Subgroup analyses and analyses conducted on the secondary endpoints (PDR, SSLDR, AADR, and hyperplastic polyps detection rate) should be considered exploratory.
Preliminarily, clinically relevant covariates balancing across the four groups were assessed. For each outcome (ADR, PDR, SSLDR, AADR, hyperplastic polyps detection rate), we calculated χ2 tests and fitted generalized linear models (GLM) with a binomial family and identity link to calculate differences in detection rates and 95%CI. To evaluate effect modification by endoscopists’ volume, we included in the model product terms (CADe × volume). Similarly, we evaluated effect modification by center and indication (screening vs diagnostic) by including the relevant product terms in the models. Continuous variables (e.g., endoscopy times) were analyzed using Wilcoxon (Mann-Whitney) tests. For selected endpoints (i.e., PDR and ADR), to evaluate effect modification by endoscopists’ volume, we performed sensitivity analyses using either generalized estimating equation (GEE) models with robust standard error or mixed effect (i.e., random-intercept or hierarchical) models again with binomial family and identity link, hereby accounting potential within-center outcome correlations.
Consistent with the literature, study analyses were conducted on the per-protocol population after the assessment of post-randomization exclusion criteria (Supplementary Figure 1); P values were reported without strict dichotomization at the 0.05 threshold, and results were interpreted in the context of effect sizes[25,26]. Analyses were performed using Stata software (StataCorp version 19, 2025).
RESULTS
Overall and inter-centers analyses
A total of 421 patients were randomized from August 2023 to September 2024, and 409 were included in the per-protocol population for final analyses, after removal of six patients per arm for inadequate bowel preparation (Boston Bowel Preparation Scale score of < 2 in any colonic segment, n = 5) and incomplete procedural or histological data (n = 7), consistent with non-resected or non-retrieved lesions (Supplementary Figure 1). Baseline characteristics were comparable in the CADe and CC group and are depicted in Table 1.
Table 1 Baseline characteristics, per protocol population, median (interquartile range)/n (%).
A total of 332 polyps were identified, of which 316 were resected and retrieved; 6 polyps were only biopsied (i.e., suspecting deep invasion), and 10 polyps were resected but not retrieved (not included for further analysis). Among them, 208 (64.6%) were adenomas, of which 7 (3.4% of adenomas) were serrated sessile adenomas, 82 (25.5%) were either hyperplastic or inflammatory polyps, and 32 (9.9%) serrated sessile polyps. The most common polyps’ site was the ascending colon (92 polyps, 28.6%), followed by the transverse colon (56 polyps, 17.8%), and the rectum (55 polyps, 17.1%).
Overall, PDR, ADR, AADR, and SSLDR were equal to 43.3% (n = 177), 30.8% (n = 126), 5.6% (n = 23), and 7.6% (n = 31), respectively. Differences were observed among the centers in terms of AADR and SSLDR, although the overall figures remained very low. Details for every center are illustrated in Table 2.
The APC rate was 0.51. The SSLs per colonoscopy rate was 0.08. We reported an overall PDR of 39.4% for primary screening, 51.7% for secondary screening, 57.7% for surveillance, and 38.9% for diagnostic. The overall ADR was equal to 28.3% for primary screening, 46.7% for secondary screening, 50% for surveillance, and 22.2% for diagnostic colonoscopies. In a screening vs. diagnostic colonoscopies sub-analysis, PDR was 47.4% vs 38.9% and ADR 38.9% vs 22.2%, respectively. Regarding daily operator fatigue, the median pre-procedure workload was 3 hours and two procedures, with an interquartile range (IQR) of 1-5 and 0-4) overall. Endoscopists at Policlinico had the highest workload (5 hours, three procedures; P = 0.0001), with no difference between CADe and CC arms.
CADe vs CC
The use of CADe was equally distributed among HVE and LVE, patients’ age, and gender (P > 0.05). Overall, we did not observe clinically relevant changes in terms of PDR (CADe 41.2%, n = 84, vs CC 45.4%, n = 93, P = 0.39), ADR (CADe 29.4%, n = 60, vs CC 32.2%, n = 66, P = 0.54), AADR (CADe 5.9%, n = 12, vs CC 5.4%, n = 11, P = 0.82), SSLDR (CADe 6.3%, n = 13, vs CC 8.7%, n = 18, P = 0.36), hyperplastic PDR (CADe 25.6%, n = 21, vs CC 23.4%, n = 21, P = 0.99). Similar results for PDR and ADR were obtained by an analysis stratified by center (Supplementary Table 1). There was no effect modification in terms of ADR and PDR by the four main indications (Table 3).
Table 3 Polyp detection rate and adenoma detection rate: Conventional colonoscopy vs computer-aided detection overall and for the main indications, n (%).
Similarly, we did not observe clinically relevant differences in SSLs per colonoscopy (CADe 0.07 vs CC 0.09, P = 0.49) or in APC (CADe 0.50 vs CC 0.52, P = 0.86). Moreover, we found that there was no effect modification by indication (i.e., screening or diagnostic colonoscopy) on either the PDR or the ADR in CADe compared to CC overall, as well as in high-volume or in low-volume operators (P-interaction ranging from 0.11 to 0.66, Figures 1 and 2). These findings were confirmed after stratification according to the daily working-related fatigue parameters. No differences were identified between the CADe and CC arms in terms of median time to reach the cecum (10 minutes, IQR 7-15, vs 10 minutes, IQR 7-14, P = 0.53) and median overall procedure time (22 minutes, IQR 18-30, vs 22 minutes, IQR 18-30, P = 0.56).
Figure 1 Polyp detection rate.
Conventional colonoscopy vs computer-aided detection overall, in low-volume endoscopists and high-volume endoscopists. P values refer to P-interaction in a generalized linear model with binomial family and identity link. PDR: Polyp detection rate; CC: Conventional colonoscopy; CADe: Computer-aided detection; LVE: Low-volume endoscopists; HVE: High-volume endoscopists.
Figure 2 Adenoma detection rate.
Conventional colonoscopy vs computer-aided detection overall, in low-volume endoscopists and high-volume endoscopists. P values refer to P-interaction in a generalized linear model with binomial family and identity link. ADR: Adenoma detection rate; CC: Conventional colonoscopy; CADe: Computer-aided detection; LVE: Low-volume endoscopists; HVE: High-volume endoscopists.
HVE vs LVE
The procedures were performed by HVE or LVE with an overall distribution of 54.5% (n = 223) vs 45.5% (n = 186), respectively. In the HVE group, the PDR, ADR, and SSLDR were 47.1%, 34.1%, and 8%, respectively. The LVE group had a PDR, ADR, and SSLDR of 38.7%, 26.9%, and 7.0% respectively. Among HVE operators, PDR was smaller in the CADe group (40.5%) compared with the CC group (54.2%, P = 0.04 with GLM). Conversely, among LVE operators PDR was similar in the CADe compared to CC group (42.1% vs 35.7%, P = 0.38 with GLM). There was no robust evidence of effect modification by operators’ volume: P-interaction = 0.04 with GLM (Figure 3, left panel), 0.10 with the GEE model, and 0.10 with the mixed model.
Figure 3 Polyp detection rate and adenoma detection rate in low-volume endoscopist vs high-volume endoscopist. P values refer to P interaction in a generalized linear model with binomial family and identity link. PDR: Polyp detection rate; ADR: Adenoma detection rate; CC: Conventional colonoscopy; CADe: Computer-aided detection; LVE: Low-volume endoscopists; HVE: High-volume endoscopists.
Similarly, among HVE operators, ADR was smaller in the CADe group compared with the CC group (29.3% vs 39.3%, P = 0.12 with GLM). Among LVE operators, ADR was similar in the CADe compared to CC group (29.6% vs 24.5%, P = 0.44 with GLM). There was no evidence of effect modification by operators’ volume: P-interaction = 0.10 with GLM (Figure 3, right panel), 0.16 with the GEE model, and 0.18 with the mixed model. Findings on PDR and ADR by study arm and operators’ volume were quite consistent across different centers (Table 4).
Table 4 Polyp detection rate and adenoma detection rate of the two study arms among high-volume and low-volume endoscopists in the different centers, n (%).
No differences were reported in terms of AADR and SSLDR with CADe compared to CC either in procedures performed by the high-volume or those performed by the low-volume operators. The hyperplastic polyps’ detection rate, based on a per-lesion analysis, was similar among HVE and LVE in the two study arms. In relation to procedural time, there was an increase in overall procedure median time among LVE using CADe compared to LVE during CC (27.5 minutes, IQR 20-35, vs 24 minutes, IQR 19-31.5, P = 0.02). On the contrary, the overall procedure time was not affected by CADe among HVE.
DISCUSSION
Polyp detection parameters
Our findings suggest that CADe systems in screening and diagnostic colonoscopies do not improve key performance metrics (PDR, ADR, SSLDR, APC SSLs per colonoscopy) when standard colonoscopy is performed using full HD endoscopic equipment and UHD video monitors. Supplementary Table 2 is a summary of RCTs from the last five years evaluating CADe systems in colonoscopy[7,18,27-43].
On CADe-assisted colonoscopy, a meta-analysis of 21 RCTs of 18232 patients found an 8.1% absolute ADR increase (44.0% vs 35.9%) and higher APC (0.90 vs 0.71) with CADe, but no improvement in AADR[17]. It also reported an increased rate of non-neoplastic polyp resections, raising concerns about CADe impact on cancer outcomes[17]. An Italian multicenter RCT (685 patients), including HVE only, found a 14.4% absolute ADR increase (54.8% vs 40.4%) with CADe (GI Genius™, Medtronic, United States), without differences in non-neoplastic lesion resection or withdrawal time[20]. However, the significant advantages observed in the early phase of AI studies were not confirmed in more recent years, and, particularly in real-world studies[44,45].
An international RCT published in 2024 (DISCOVERY, PENTAX Medical, Tokyo, Japan) found similar ADR (38.4% vs 37.7%; P = 0.43) and PDR (55.1% vs 50.7%; P = 0.29) between CADe and CC[33]. The results are comparable to many other real-life, controlled trials and a systematic review of 9782 patients in real-world settings[46]. Notably, in the study by Maas et al[33], CADe was associated with a small, albeit clinically relevant, increase in SSLs per colonoscopy rate (0.30 vs 0.19, P = 0.049)[33] not mirrored in our study. This is possibly due to a higher SSLs prevalence, to the well-described variability in histopathological SSLs classification, and to important differences existing between the two study designs (i.e., stratification by operators’ volume, high proportion of diagnostic colonoscopies, UHD environment leading to a potential ceiling effect).
In a larger multicenter RCT on 1627 patients undergoing screening or diagnostic colonoscopy in 12 German institutions published in 2025, another CADe system (Fujifilm Europe, Ratingen, Germany) did not improve ADR (40% vs 37.5%, CADe+ vs CADe-), causing a slight increase in resected hyperplastic polyps (odds ratio = 1.3; 95%CI: 1.0-1.6)[44]. A non-randomized study on the Endo-Aid CADe AI system (Olympus, Japan) with UHD monitors found no ADR, PDR, or SSLDR differences[21], aligning with our findings. Similarly, in our study, procedures in both arms were performed using UHD monitors. In contrast, many studies lack clarity on monitor quality, raising the possibility of technology bias. Potential impacts of introducing CADe into UHD rather than normal definition monitors should be further investigated. Additionally, analyzing individual endoscopists’ ADR, PDR, and SSLDR rates in CADe vs CC could clarify the CADe impact. Comparing pre-trial and trial detection rates may reveal a Hawthorne effect, where physicians change behavior under observation during a study[47].
Screening vs diagnostic colonoscopies
Our detection rates across the four main colonoscopy indications align with Fernandes’ meta-analysis, which reported an overall ADR of 27% (27% in primary screening, 34% in secondary screening, 43% in surveillance, and 25% in diagnostic procedures)[48]. Our study showed higher overall and screening ADRs, with no difference between CADe and standard procedures. However, CADe showed a slightly reduced performance in diagnostic procedures, especially among HVE. A recent RCT by Maas et al[33] also reported slight ADR discrepancies by indication, with a slight increase in diagnostic and surveillance colonoscopies but a 7.7% ADR reduction in non-fecal occult blood test screening (P = 0.45). Nearly half of our cohort underwent diagnostic colonoscopies, potentially influencing overall results. These findings highlight the need to consider procedural indications when evaluating CADe, suggesting its effectiveness may be context dependent.
HVE vs LVE
In our study, CADe provided a modest, statistically non-significant benefit for LVE, likely aiding in detecting subtle lesions missed due to limited experience or suboptimal mucosal exposure. This may align with Hassan et al’s meta-analysis[17], which found CADe more effective in settings with lower experience or baseline detection rates. On the other hand, in our study the use of CADe resulted in a trend to lower ADR and, particularly, PDR in the HVE group, especially during diagnostic colonoscopy. This apparent counterintuitive outcome may be explained by overreliance on the system, leading to a form of cognitive disengagement or reduced vigilance when CADe is employed. High-volume operators, confident in their skills, may unintentionally deprioritize active scanning, trusting the system to compensate, which paradoxically undermines its effectiveness[49]. A recent study by Troya et al[50] showed that the use of CADe systems increases misinterpretation of normal mucosa and decreases the endoscopists’ eye travel distance, laying the foundations for a progressive deskilling.
In addition, there are some perceptual and cognitive bottlenecks which could have an important role in errors during colonoscopies, as described by Introzzi et al[51]. The process of identifying mucosal lesions during a colonoscopy exemplifies a guided search strategy, which enables the detection of complex patterns. However, this approach can be detrimental to accuracy, as it may lead to inattentional blindness, increasing the likelihood of missing lesions. Additionally, the over-alternation bias can cause operators to believe that the same event is unlikely to occur again shortly after it has already been encountered, leading to higher miss rates following the first polypectomy. Another cognitive bias is overconfidence, where endoscopists’ self-reported accuracy estimates tend to be higher than their actual performance. This overconfidence becomes problematic if inaccurate judgments are not corrected through feedback[52]. In this context, AI has the potential to play a pivotal role by supporting endoscopists, especially less experienced ones, through real-time feedback. However, the human-AI interaction can be influenced by ego bias, leading to an implicit and unwarranted preference for one’s own opinions. Conversely, endoscopists may overestimate the capabilities of AI, which can result in reduced attention during procedures and, unconsciously, a decline in the quality of mucosal exposure[19]. In LVE, the overall procedure time was slightly longer in the CADe compared to the CC group (27 minutes vs 24 minutes), likely due to increased visual stimulation and false positives, consistent with existing literature[17].
Operator fatigue
Our analysis found no correlation between operator fatigue measured by hours worked or procedure volume and detection metrics. While studies suggest fatigue may impair the guided search process in colonoscopy[51], tertiary referral centers likely provide a structured workload that mitigates its impact. Fatigue was assessed using indirect workload-based proxies rather than validated fatigue scales, limiting the precision of effect modification analyses: Our findings should therefore be considered hypothesis-generating. Different experimental settings and more precise fatigue parameters may better reveal the effects on performance.
Limits of the study
Among the limitations of the study, it should be noted that it was conducted in tertiary care centers. This may have minimized the potential benefit of CADe, narrowing the generalizability of the results to community hospitals or lower-resource environments. The study sample size was underpowered to trace conclusive analyses on operator fatigue burden, learning or training effect (e.g., variable ADR differences between initial and late study phase). In addition, the current literature lacks solid evidence based on standardized measures with adequate construct validity, leading interpretation of effect modification by “fatigue” or “learning effect” quite challenging.
A few recruited patients were not included in the per-protocol population based on exclusion criteria. However, their absolute and relative number was very small (12/421) with a negligible impact on study results. Randomization was performed 1:1 via REDCap without stratification by site, or by endoscopist, or by indication. Endoscopists were necessarily aware of CADe use, therefore potentially introducing performance bias resulting from enhancing withdrawal techniques, inspection intensity, and polypectomy decisions. Baseline ADR was not available for all endoscopists, hampering comparative analyses with CADe exposure. Prior exposure of endoscopists to CADe was not systematically investigated, hampering post hoc analyses on deskilling or human-AI interaction bias. Finally, although sensitivity analyses using GEE and mixed models were conducted for selected endpoints to address non-independence, these approaches primarily accounted for center-level correlation and did not explicitly model clustering at the level of the individual endoscopist. This may have resulted in residual within-operator correlation.
CONCLUSION
While CADe systems show promise as supportive tools in colonoscopy, their effectiveness may differ depending on operator experience, procedural context, and application strategies. Our RCT was designed including UHD monitors in both arms to avoid technological bias, and revealed a statistically non-significant, moderate improvement in ADR only among LVE. Conversely, HVE did not improve their detection metrics, especially during diagnostic colonoscopies compared to screening procedures. This suggests that unintended consequences, such as overreliance on technology, may counteract potential benefits of CADe. To mitigate this, tailored training programs and refined implementation strategies might be essential to optimize collaboration between endoscopists and CADe systems[52]. Future research should further investigate different human-AI interaction strategies, focusing on how operator experience and procedural context influence outcomes to ensure CADe systems are applied effectively and equitably across diverse clinical settings.
Levin B, Lieberman DA, McFarland B, Andrews KS, Brooks D, Bond J, Dash C, Giardiello FM, Glick S, Johnson D, Johnson CD, Levin TR, Pickhardt PJ, Rex DK, Smith RA, Thorson A, Winawer SJ; American Cancer Society Colorectal Cancer Advisory Group; US Multi-Society Task Force; American College of Radiology Colon Cancer Committee. Screening and surveillance for the early detection of colorectal cancer and adenomatous polyps, 2008: a joint guideline from the American Cancer Society, the US Multi-Society Task Force on Colorectal Cancer, and the American College of Radiology.Gastroenterology. 2008;134:1570-1595.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 1554][Cited by in RCA: 1428][Article Influence: 79.3][Reference Citation Analysis (5)]
Repici A, Spadaccini M, Antonelli G, Correale L, Maselli R, Galtieri PA, Pellegatta G, Capogreco A, Milluzzo SM, Lollo G, Di Paolo D, Badalamenti M, Ferrara E, Fugazza A, Carrara S, Anderloni A, Rondonotti E, Amato A, De Gottardi A, Spada C, Radaelli F, Savevski V, Wallace MB, Sharma P, Rösch T, Hassan C. Artificial intelligence and colonoscopy experience: lessons from two randomised trials.Gut. 2022;71:757-765.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 210][Cited by in RCA: 175][Article Influence: 43.8][Reference Citation Analysis (12)]
Rees CJ, Thomas Gibson S, Rutter MD, Baragwanath P, Pullan R, Feeney M, Haslam N; British Society of Gastroenterology, the Joint Advisory Group on GI Endoscopy, the Association of Coloproctology of Great Britain and Ireland. UK key performance indicators and quality assurance standards for colonoscopy.Gut. 2016;65:1923-1929.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 250][Cited by in RCA: 244][Article Influence: 24.4][Reference Citation Analysis (4)]
Kaminski MF, Thomas-Gibson S, Bugajski M, Bretthauer M, Rees CJ, Dekker E, Hoff G, Jover R, Suchanek S, Ferlitsch M, Anderson J, Roesch T, Hultcranz R, Racz I, Kuipers EJ, Garborg K, East JE, Rupinski M, Seip B, Bennett C, Senore C, Minozzi S, Bisschops R, Domagk D, Valori R, Spada C, Hassan C, Dinis-Ribeiro M, Rutter MD. Performance measures for lower gastrointestinal endoscopy: a European Society of Gastrointestinal Endoscopy (ESGE) Quality Improvement Initiative.Endoscopy. 2017;49:378-397.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 591][Cited by in RCA: 542][Article Influence: 60.2][Reference Citation Analysis (5)]
Bisschops R, East JE, Hassan C, Hazewinkel Y, Kamiński MF, Neumann H, Pellisé M, Antonelli G, Bustamante Balen M, Coron E, Cortas G, Iacucci M, Yuichi M, Longcroft-Wheaton G, Mouzyka S, Pilonis N, Puig I, van Hooft JE, Dekker E. Advanced imaging for detection and differentiation of colorectal neoplasia: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - Update 2019.Endoscopy. 2019;51:1155-1179.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 291][Cited by in RCA: 257][Article Influence: 36.7][Reference Citation Analysis (5)]
Kröner PT, Engels MM, Glicksberg BS, Johnson KW, Mzaik O, van Hooft JE, Wallace MB, El-Serag HB, Krittanawong C. Artificial intelligence in gastroenterology: A state-of-the-art review.World J Gastroenterol. 2021;27:6794-6824.
[PubMed] [DOI] [Full Text]
Hassan C, Spadaccini M, Mori Y, Foroutan F, Facciorusso A, Gkolfakis P, Tziatzios G, Triantafyllou K, Antonelli G, Khalaf K, Rizkala T, Vandvik PO, Fugazza A, Rondonotti E, Glissen-Brown JR, Kamba S, Maida M, Correale L, Bhandari P, Jover R, Sharma P, Rex DK, Repici A. Real-Time Computer-Aided Detection of Colorectal Neoplasia During Colonoscopy : A Systematic Review and Meta-analysis.Ann Intern Med. 2023;176:1209-1220.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 177][Cited by in RCA: 160][Article Influence: 53.3][Reference Citation Analysis (3)]
Repici A, Badalamenti M, Maselli R, Correale L, Radaelli F, Rondonotti E, Ferrara E, Spadaccini M, Alkandari A, Fugazza A, Anderloni A, Galtieri PA, Pellegatta G, Carrara S, Di Leo M, Craviotto V, Lamonaca L, Lorenzetti R, Andrealli A, Antonelli G, Wallace M, Sharma P, Rosch T, Hassan C. Efficacy of Real-Time Computer-Aided Detection of Colorectal Neoplasia in a Randomized Trial.Gastroenterology. 2020;159:512-520.e7.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 551][Cited by in RCA: 469][Article Influence: 78.2][Reference Citation Analysis (12)]
Veitch AM, Radaelli F, Alikhan R, Dumonceau JM, Eaton D, Jerrome J, Lester W, Nylander D, Thoufeeq M, Vanbiervliet G, Wilkinson JR, Van Hooft JE. Endoscopy in patients on antiplatelet or anticoagulant therapy: British Society of Gastroenterology (BSG) and European Society of Gastrointestinal Endoscopy (ESGE) guideline update.Gut. 2021;70:1611-1628.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 168][Cited by in RCA: 148][Article Influence: 29.6][Reference Citation Analysis (5)]
Spada C, Cesaro P, Fuccio L, Salvi D, Ferrari C, Barbaro F, Bizzotto A, Butitta F, Gerardi V, Lovera M, Milluzzo SM, Minelli Grazioli L, Olivari N, Pecere S, Piccirelli S, Pugliano CL, Pesatori EV, Quadarella A, Tettoni E, Zani C, Ricciardiello L, Costamagna G. Impact of Artificial Intelligence for Detection of Precancerous Colonic Lesions in a Fecal Immunochemical Blood Test-Based Organized Screening Program in Italy: A Randomized Control Trial.United European Gastroenterol J. 2026;14:e70176.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 1][Reference Citation Analysis (0)]
Jeon HJ, Keum B, Jeong ES, Kim SE, Moon CM, Lee B, Kim S, Choi HS, Lee JM, Kim ES, Jeen YT. Clinical Efficacy of Real-Time Artificial Intelligence-Assisted Colonoscopy in Colorectal Polyp Detection: A Prospective Multicenter Randomized Controlled Trial.Gut Liver. 2026;20:97-106.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 1][Reference Citation Analysis (0)]
Xu X, Ba L, Lin L, Song Y, Zhao C, Yao S, Cao H, Chen X, Mu J, Yang L, Feng Y, Wang Y, Wang B, Zheng Z. Evaluation efficacy and accuracy of a real-time computer-aided polyp detection system during colonoscopy: a prospective, multicentric, randomized, parallel-controlled study trial.Surg Endosc. 2025;39:7417-7427.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 4][Reference Citation Analysis (0)]
Yabuuchi Y, Hosotani K, Morihisa Y, Fujio Y, Oshikawa D, Oshita M, Iketani M, Tsukamoto K, Sone A, Nanjo T, Tatsuno R, Tanaka K, Nagao S, Akiyama S, Tanke G, Wada M, Morita S, Inoue S, Chung H, Nishikawa Y, Inokuma T. Effect of Computer-Aided Detection During Colonoscopy on Adenoma Detection Rate in a Community Hospital Setting: Randomized Controlled Trial.Dig Endosc. 2025;37:1215-1223.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 1][Cited by in RCA: 6][Article Influence: 6.0][Reference Citation Analysis (0)]
Djinbachian R, Taghiakbari M, Barkun A, Medawar E, Alj A, Sidani S, Kiow JLC, Panzini B, Bouin M, Lasfar D, von Renteln D. Optimized computer-assisted technique for increasing adenoma detection during colonoscopy: a randomized controlled trial.Surg Endosc. 2025;39:1120-1127.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 7][Reference Citation Analysis (0)]
Maas MHJ, Rath T, Spada C, Soons E, Forbes N, Kashin S, Cesaro P, Eickhoff A, Vanbiervliet G, Salvi D, Belletrutti PJ, Siersema PD; Discovery study team. A computer-aided detection system in the everyday setting of diagnostic, screening, and surveillance colonoscopy: an international, randomized trial.Endoscopy. 2024;56:843-850.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 11][Cited by in RCA: 11][Article Influence: 5.5][Reference Citation Analysis (0)]
Thiruvengadam NR, Solaimani P, Shrestha M, Buller S, Carson R, Reyes-Garcia B, Gnass RD, Wang B, Albasha N, Leonor P, Saumoy M, Coimbra R, Tabuenca A, Srikureja W, Serrao S. The Efficacy of Real-time Computer-aided Detection of Colonic Neoplasia in Community Practice: A Pragmatic Randomized Controlled Trial.Clin Gastroenterol Hepatol. 2024;22:2221-2230.e15.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 22][Cited by in RCA: 18][Article Influence: 9.0][Reference Citation Analysis (0)]
Ahmad A, Wilson A, Haycock A, Humphries A, Monahan K, Suzuki N, Thomas-Gibson S, Vance M, Bassett P, Thiruvilangam K, Dhillon A, Saunders BP. Evaluation of a real-time computer-aided polyp detection system during screening colonoscopy: AI-DETECT study.Endoscopy. 2023;55:313-319.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 70][Cited by in RCA: 67][Article Influence: 22.3][Reference Citation Analysis (0)]
Gimeno-García AZ, Hernández Negrin D, Hernández A, Nicolás-Pérez D, Rodríguez E, Montesdeoca C, Alarcon O, Romero R, Baute Dorta JL, Cedrés Y, Castillo RD, Jiménez A, Felipe V, Morales D, Ortega J, Reygosa C, Quintero E, Hernández-Guerra M. Usefulness of a novel computer-aided detection system for colorectal neoplasia: a randomized controlled trial.Gastrointest Endosc. 2023;97:528-536.e1.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 59][Cited by in RCA: 57][Article Influence: 19.0][Reference Citation Analysis (1)]
Shaukat A, Lichtenstein DR, Somers SC, Chung DC, Perdue DG, Gopal M, Colucci DR, Phillips SA, Marka NA, Church TR, Brugge WR; SKOUT™ Registration Study Team. Computer-Aided Detection Improves Adenomas per Colonoscopy for Screening and Surveillance Colonoscopy: A Randomized Trial.Gastroenterology. 2022;163:732-741.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 136][Cited by in RCA: 110][Article Influence: 27.5][Reference Citation Analysis (4)]
Rondonotti E, Di Paolo D, Rizzotto ER, Alvisi C, Buscarini E, Spadaccini M, Tamanini G, Paggi S, Amato A, Scardino G, Romeo S, Alicante S, Ancona F, Guido E, Marzo V, Chicco F, Agazzi S, Rosa C, Correale L, Repici A, Hassan C, Radaelli F; AIFIT Study Group. Efficacy of a computer-aided detection system in a fecal immunochemical test-based organized colorectal cancer screening program: a randomized controlled trial (AIFIT study).Endoscopy. 2022;54:1171-1179.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 60][Cited by in RCA: 58][Article Influence: 14.5][Reference Citation Analysis (4)]
Yao L, Zhang L, Liu J, Zhou W, He C, Zhang J, Wu L, Wang H, Xu Y, Gong D, Xu M, Li X, Bai Y, Gong R, Sharma P, Yu H. Effect of an artificial intelligence-based quality improvement system on efficacy of a computer-aided detection system in colonoscopy: a four-group parallel study.Endoscopy. 2022;54:757-768.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 105][Cited by in RCA: 91][Article Influence: 22.8][Reference Citation Analysis (3)]
Zimmermann-Fraedrich K, Sehner S, Rösch T, Aschenbeck J, Schubert S, Liceni T, Moog G, Neumann H, Berndt R, Weigt J, Kaczmarek DJ, May A, Hoffmeister A, Möschler O, Wiessner C, Schachschal G. No Effect of Computer-Aided Diagnosis on Colonoscopic Adenoma Detection in a Large Pragmatic Multicenter Randomized Study.Am J Gastroenterol. 2026;121:881-889.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 5][Cited by in RCA: 6][Article Influence: 6.0][Reference Citation Analysis (0)]
Bretthauer M, Ahmed J, Antonelli G, Beaumont H, Beg S, Benson A, Bisschops R, De Cristofaro E, Gibbons E, Häfner M, Karsenti D, Laquière A, Loly JP, O'Reilly SM, Pellisé M, Grubelic Ravic K, Triantafyllou K, Tziatzios G, Valente R, Walter BM, Wiesand M, Lorenzo-Zúñiga V, Gralnek IM. Use of computer-assisted detection (CADe) colonoscopy in colorectal cancer screening and surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement.Endoscopy. 2025;57:667-673.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 25][Cited by in RCA: 21][Article Influence: 21.0][Reference Citation Analysis (0)]
Patel HK, Mori Y, Hassan C, Rizkala T, Radadiya DK, Nathani P, Srinivasan S, Misawa M, Maselli R, Antonelli G, Spadaccini M, Facciorusso A, Khalaf K, Lanza D, Bonanno G, Rex DK, Repici A, Sharma P. Lack of Effectiveness of Computer Aided Detection for Colorectal Neoplasia: A Systematic Review and Meta-Analysis of Nonrandomized Studies.Clin Gastroenterol Hepatol. 2024;22:971-980.e15.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 72][Cited by in RCA: 59][Article Influence: 29.5][Reference Citation Analysis (0)]
Budzyń K, Romańczyk M, Kitala D, Kołodziej P, Bugajski M, Adami HO, Blom J, Buszkiewicz M, Halvorsen N, Hassan C, Romańczyk T, Holme Ø, Jarus K, Fielding S, Kunar M, Pellise M, Pilonis N, Kamiński MF, Kalager M, Bretthauer M, Mori Y. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study.Lancet Gastroenterol Hepatol. 2025;10:896-903.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 5][Cited by in RCA: 85][Article Influence: 85.0][Reference Citation Analysis (0)]
Cabitza F, Campagner A, Fregosi C, Cameli M, Gallazzi E, Sconfienza LM, Tontini GE. Five Degrees of Separation: Investigating the Unexpected Potential of Displaced Human-AI Collaboration Protocols for Apter AI Support.Proc ACM Hum -Comput Interact. 2025;9:1-28.
[PubMed] [DOI] [Full Text]
Footnotes
Peer review: Externally peer reviewed.
Peer-review model: Single blind
Specialty type: Gastroenterology and hepatology
Country of origin: Italy
Peer-review report’s classification
Scientific quality: Grade B, Grade B, Grade D
Novelty: Grade B, Grade C, Grade D
Creativity or innovation: Grade B, Grade C, Grade D
Scientific significance: Grade B, Grade C, Grade D
P-Reviewer: Ichita C, Chief, MD, Japan; Mihara H, Associate Professor, MD, PhD, Japan S-Editor: Wu S L-Editor: A P-Editor: Wang CH
