Effect of Combination of a Mucosal Exposure Device and Computer-Aided Detection in Diagnostic, Screening, and Surveillance Colonoscopy: An International, Multicenter Study

Abstract

INTRODUCTION:

Mucosal exposure devices (MEDs) and computer-aided detection (CADe) systems may both improve adenoma detection through distinct mechanisms: expanding mucosal visualization and highlighting lesions, respectively. This study investigated the efficacy of combining CADe-assisted colonoscopy with a MED compared with CADe-assisted colonoscopy alone.

METHODS:

This international, multicenter, prospective, nonrandomized, single-arm study (NTC05220345) was conducted at 3 centers that also participated in the previous DISCOVERY II randomized controlled trial, comparing CADe-assisted with conventional colonoscopy. Eligible participants were patients referred for diagnostic, nonfecal immunochemical test screening, or surveillance colonoscopy and underwent CADe-assisted colonoscopy (DISCOVERY, PENTAX Medical) with a MED using an integrated inflatable balloon (G-EYE, SMART Medical Systems). The primary outcome was adenoma detection rate (ADR); secondary outcomes included sessile serrated lesion detection rate and withdrawal time without interventions. Outcomes were compared with historical controls of the CADe-arm of the DISCOVERY II study.

RESULTS:

Of 196 enrolled participants, 182 were included in the final analysis and compared with 250 participants from the historical CADe-arm. ADR was 47.3% in the CADe + MED-group vs 38.4% in the CADe-group (P = .066; absolute difference: 8.9%, 95% confidence interval: 0.6–18.3). Mixed-effects logistic regression model adjusting for clustering and confounders calculated an odds ratio of 1.16 (95% confidence interval: 0.74–1.81). Median withdrawal time was slightly longer with CADe + MED compared with CADe-only (10.0 vs 9.2 minutes, P = 0.004), whereas sessile serrated lesion detection rate was not significantly different (12.6% vs 18.4%, P = 0.11).

DISCUSSION:

In this study using historical controls, CADe-assisted colonoscopy combined with a MED did not significantly increase ADR compared with CADe alone, suggesting limited synergistic benefit.

INTRODUCTION

Although colonoscopy reduces the incidence of colorectal cancer (CRC) (1), a substantial miss rate of up to 34% for colonic neoplasia remains (2,3). Missed lesions are hypothesized to account for a considerable proportion of postcolonoscopy CRCs (4,5). Colonoscopy results are partly operator-dependent (6). Therefore, factors such as temporary inattentiveness and inadequate mucosal visualization, particularly behind haustral folds, are thought to be contributing factors for lesions being missed (7,8).

Mucosal exposure devices (MEDs), such as the G-EYE endoscope (SMART Medical Systems) with an integrated inflatable balloon, improve visualization by mechanically flattening these haustral folds, thereby increasing detection rates and reduce adenoma miss rates compared with conventional colonoscopy (9–11). Computer-aided detection (CADe) has also been shown to increase adenoma detection rate (ADR) and other detection metrics (12,13). CADe highlights suspected lesions on the endoscopic monitor, but does not improve mucosal exposure and remains dependent on the endoscopist to visualize a lesion before it can be detected.

Given these complementary mechanisms, a synergistic effect of combining CADe with MEDs has been proposed. This hypothesis is supported by a post hoc video analysis of a previous tandem CADe study, in which many missed lesions were outside the visual field and thus undetectable by CADe alone (14). However, recent studies evaluating this combined approach have reported mixed results regarding incremental benefit (15–17).

The aim of this study was to evaluate the effect of CADe-assisted colonoscopy combined with a MED on ADR, compared with the CADe-assisted colonoscopy arm of a previously conducted randomized controlled trial (RCT) (18), in patients undergoing diagnostic, nonfecal immunochemical test (FIT) screening, or surveillance colonoscopy.

METHODS

Study design and participants

This international, multicenter, prospective, interventional, single-arm study was conducted at 3 hospitals in Germany, Italy, and The Netherlands. The results were compared with the findings from the CADe-assisted arm of the previously conducted DISOVERY II RCT, which compared conventional colonoscopy to CADe-assisted colonoscopy for ADR, adenoma per colonoscopy, and detection of sessile serrated lesions (SSLs) (18). The current study included the 3 highest-enrolling centers from the DISCOVERY II study (n = 155/250, 62%), which were among the 7 centers participating in that study.

Eligible participants, aged 18 years and older, were referred for non-FIT screening, surveillance, or diagnostic colonoscopy (excluding referrals based on positive FIT). Patients were included on a consecutive base. The inclusion criteria were identical to the DISCOVERY II study (18). Exclusion criteria were nearly identical, except for an additional exclusion criterion mandated by the institutional review board of the coordinating site, excluding patients with previous surgical resection of any portion of the colon. Additional exclusion criteria are reported in the Supplementary Digital Content (see Supplementary Methods, http://links.lww.com/CTG/B439).

The study was registered at ClinicalTrials.gov (identifier: NCT05220345), approved by independent institutional review boards at each site and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All participants provided written informed consent. Data verification and on-site monitoring of all 3 sites was conducted by an independent clinical research monitor. The study was reported according to the Consolidated Standards of Reporting Trials - Artificial Intelligence guidelines. All authors had access to the data and approved the final manuscript.

Computer aided detection system

All colonoscopies were performed using the DISCOVERY system (PENTAX Medical, Tokyo, Japan) with software versions 1.0.4. The system detects suspected polyps in real-time and displays a bounding box on the endoscopic monitor to alert the endoscopist (Figure 1). Additional details of this system have been described previously (18).

Figure 1.

Detection by the CADe system. CADe generated overlay of a blue bounding box highlighting a lesion during real-time colonoscopy. CADe, computer-aided detection.

Mucosal exposure device

The G-EYE endoscope is a standard colonoscope with a permanently integrated, inflatable balloon on the distal bending section (Figure 2). When deflated, the outer diameter of the colonoscope is approximately 0.2 mm greater than that of a conventional colonoscope. The balloon is pressure-controlled and can be inflated, up to 60 mm in diameter, by a foot pedal or handheld remote. The system automatically deflates depending on the applied pressure, allowing advancement through more narrow luminal segments. Balloon inflation during withdrawal flattens the haustral folds, stabilizes the endoscope, and reduces bowel slippage. A previous study reported no increase in adverse events with this MED compared with conventional colonoscopy (19).

Figure 2.

The G-Eye system. The mucosal exposure device (MED) with the integrated balloon inflated. MED, mucosal exposure device.

Study investigators

The study was performed by 8 experienced endoscopists, 7 of whom also participated in the CADe arm of the DISCOVERY II study. All endoscopists underwent standardized training in the use of the CADe system and the MED and completed at least 3 combined training procedures before patient enrollment. Endoscopist eligibility required a minimum of 500 independently performed colonoscopies per endoscopist.

Study procedures

The CADe system was activated at the start of each procedure and remained activated during the withdrawal phase. The MED was inflated on reaching the cecum and remained inflated during the withdrawal phase. Endoscopists were instructed to maintain a withdrawal time between 6 and 10 minutes (excluding time spent on interventions) to standardize inspection time and minimize observation-related bias. Rectal diminutive polyps (1–5 mm) assessed to be hyperplastic were left in situ, based on endoscopist discretion and standard of care. Procedural aspects beyond the specified study instructions adhered to local protocols, and routine re-examination of the proximal colon was not performed. All other lesions were collected in separate containers for histopathological examination. Histopathological diagnosis was determined according to the Vienna classification (20).

Study outcomes

The primary outcome was ADR, defined as the proportion of colonoscopies with at least one histologically confirmed detected adenoma. ADR was evaluated per colonoscopy indication, endoscopist, and per study site. Secondary outcomes included mean number of adenoma per colonoscopy (total number of histologically confirmed adenomas divided by the total number of colonoscopies), polyp detection rate (PDR; proportion of colonoscopies with at least one histologically confirmed detected polyp), polyps per colonoscopy (total number of histologically confirmed polyps divided by the total number of colonoscopies), sessile serrated lesions per colonoscopy (total number of histologically confirmed SSLs divided by the total number of colonoscopies), and SSL detection rate (proportion of colonoscopies with at least one histologically confirmed detected SSL), withdrawal time (in minutes) without interventions, and total procedure time (in minutes). Index colonoscopy was defined as the first lifetime colonoscopy of a participant.

Sample size

This study was powered to detect a significant difference in ADR between CADe-assisted colonoscopy with a MED and CADe-assisted colonoscopy alone. Because this study was conceptualized before completion of the DISCOVERY II study, the initial sample size calculation was later revised based on external data from that study. This amendment was prespecified in the protocol and approved by all participating sites before analysis.

The initial sample size calculation assumed an ADR of 31% for CADe-assisted and 45% for CADe + MED procedures based on a previously observed ADR increase of approximately 10%, and an expected synergistic effect of MED when combined with CADe, yielding an assumed difference of 14 percentage points (11), resulting in a required total sample size of 418 participants (including a 10% dropout). Using a 0.6 allocation ratio due to the availability of a large CADe group from the Discovery II study, this corresponded to 276 participants in the CADe-only group and 166 in the MED + CADe group.

Following completion of the DISCOVERY II study, the sample size calculation was amended based on the observed ADR of 38.4% in the CADe-arm (18). Using a χ² test with a 2-sided alpha of 0.05 and 80% power resulted in a required total sample size of 420 participants. To maintain a comparable allocation ratio with the sample size of the historical CADe controls, a 0.7 allocation ratio was applied, resulting in a calculated sample size of 173 participants in the CADe + MED group. Accounting for an anticipated dropout rate of 10%, the final sample size for this study was set at 193 participants.

The sample size calculation was performed using G*Power, version 3.1.9.7 (Heinrich-Heine-Universität, Düsseldorf, Germany).

Statistical analysis

All analyses followed a modified intention-to-treat analysis, excluding participants postrandomization due to nonevaluable outcomes (inadequate bowel preparation [Boston Bowel Preparation Score <6] or incomplete colonoscopy). Analysis of the primary outcome was performed using a χ² test and a binary logistic regression mixed model to account for clustering and measured confounders due to the nonrandomized design of our study. Study site was included as the clustering variable, with sex, index colonoscopy, colonoscopy indication, age, and withdrawal time as confounding variables, based on protocol and uneven distribution across both groups. Clustering at the endoscopist level was explored but resulted in model instability; therefore, clustering was performed at study-site level. A mixed-effects model was used to estimate the adjusted effect of MED addition to CADe, this approach allowed direct modeling of the exposure effect while accounting for clustering and confounding. Propensity score matching was not applied, as measured confounders were directly adjusted for in the mixed-effects model. Given that both groups were derived from centers with identical inclusion criteria additional matching would not have provided additional benefit and would have reduced sample size. An exploratory subgroup analysis was performed to compare ADR across the 3 overlapping study sites. Given the exploratory nature of subgroup analysis, no correction for multiple testing was applied.

Continuous variables were presented as means (SD) or medians (interquartile range), and categorical variables as absolute numbers and percentages. Between-group comparisons were performed using independent t-tests, Mann-Whitney U tests, or χ² tests, as appropriate. The Wilson score method was used to calculate 95% confidence intervals where applicable. A 2-tailed P value < 0.05 was considered statistically significant. Cases with missing data for variables included in the mixed-effects models were excluded listwise from those analyses. Multiple imputation was not performed, as the proportion of missing data was <10%, consistent to study protocol.

Analyses were performed with Statistical Package for Social Sciences program, version 29 (IBM, Armonk, NY) and R Statistical Software (v.4.1.3; R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

The study was conducted between May 12, 2022, and August 26, 2024, during which 196 patients were enrolled to undergo CADe + MED-assisted colonoscopy. Fourteen patients were excluded, resulting in 182 participants included in the final analysis (Figure 3).

Figure 3.

Study flowchart. Other exclusions in the CADe + MED arm: incomplete colonoscopy (n = 4). BBPS, Boston bowel preparation score; CADe, computer-aided detection; MED, mucosal exposure device.

The CADe + MED group included a higher proportion of male participants compared with the CADe group (59%, n = 108/182 vs 44%, n = 109/250; P = 0.001). Colonoscopy indication also differed between groups: non-FIT screening (16%, 29/182 vs 20%, n = 50/250, respectively), surveillance (57%, 103/182 vs 39%, n = 98/250, respectively), and diagnostic (28%, n = 50/182 vs 41%, n = 102/250, respectively); P = 0.001. A lower proportion of participants underwent an index colonoscopy in the CADe + MED group compared with the CADe group (32%, 58/182 vs 45%, 112/250; P = 0.007). Additional baseline characteristics are reported in Table 1.

Table 1.

Baseline characteristics

	CADe+ mucosal exposure device (N = 182)	CADe (N = 250)	P value
Age, yr	63.0 (52.0–72.0)	60.5 (52.0–69.0)	0.18
Sex			0.001
Female	74/182 (40.7%)	141/250 (56.4%)
Male	108/182 (59.3%)	109/250 (43.6%)
Colonoscopy indication			0.001
Screening (non-FIT)	29/182 (15.9%)	50/250 (20.0%)
Surveillance	103/182 (56.6%)	98/250 (39.2%)
Diagnostic	50/182 (27.5%)	102/250 (40.8%)
Index colonoscopy, yes	58/182 (31.9%)	112/250 (44.8%)	0.007
Smoking, yes	13/182 (7.1%)	26/250 (10.4%)	0.24
Family history of CRC, yes	45/182 (24.7%)	60/250 (24.0%)	0.86
BMI, kg/m2	25.5 (23.2–28.5)	25.0 (22.5–28.8)	0.54
Total BBPSa	9.0 (6.0–9.0)	9.0 (7.0–9.0)	0.78

Data are n/N (%), or median (IQR). Bold indicates P < 0.05.

BBPS, Boston bowel preparation score; BMI, body mass index; CADe, computer-aided detection; CRC, colorectal cancer; FIT, fecal immunochemical test; IQR, interquartile range.

^a4 missed data from 1 or more segment due to incomplete colonoscopy.

Findings

The observed ADR was numerically higher in the CADe + MED group compared with CADe-assisted, although this observed trend did not reach statistical significance (47.3% vs 38.4%, P = 0.066; colonoscopies with ≥1 adenoma: n = 86/182 vs n = 96/250). A binary logistic mixed-effects model yielded an odds ratio of 1.16 (95% confidence interval [CI]: 0.74–1.81) for CADe + MED relative to CADe.

The observed sessile serrated lesions per colonoscopy was significantly lower in the CADe + MED group compared with CADe (0.15 vs 0.30, P = 0.012; total detected SSLs: 28/182 vs 76/250). However, this association was not statistically significant in the mixed model analysis (effect ratio 0.75 [95% CI: 0.49–1.15]). Similarly, SDR was not significantly different in the CADe + MED group compared with CADe (12.6% vs 18.4%, P = 0.106; total colonoscopies with ≥1 SSL: 23/182 vs 46/250). The mixed-effects model for SDR yielded an odds ratio of 0.67 [95% CI: 0.37–1.21] for CADe + MED relative to CADe. Median withdrawal time was significantly higher in the CADe + MED group compared with the CADe group (withdrawal time without interventions [interquartile range] of 10.0 [9.0–12.0] vs 9.2 [8.0–11.0], P = 0.004). Additional outcomes are reported in Tables 2 and 3, and Supplementary Digital Content (see Supplementary Table 1, http://links.lww.com/CTG/B435).

Table 2.

Primary and secondary outcomes in the modified intention-to-treat population

	CADe+ mucosal exposure device (N = 182)	CADe (N = 250)	Difference (treatment–control)	P value
Adenoma detection rate (ADR)	86/182 (47.3%)	96/250 (38.4%)	8.9 [−0.6 to 18.3]	0.066
Adenoma per colonoscopy (APC)	160/182 (0.88)	166/250 (0.66)	0.22 [−0.02 to 0.46]	0.083
Polyp detection rate (PDR)	101/182 (55.5%)	138/250 (55.2%)	0.3 [−9.2 to 9.8]	0.95
Polyps per colonoscopy (PPC)	211/182 (1.16)	299/250 (1.20)	−0.04 [−0.34 to 0.27]	0.81
Sessile serrated lesion detection rate (SDR)	23/182 (12.6%)	46/250 (18.4%)	5.8 [−1.1 to 12.6]	0.106
Sessile serrated lesions per colonoscopy (SSLPC)	28/182 (0.15)	76/250 (0.30)	−0.15 [−0.27 to −0.03]	0.012
Withdrawal time without interventions (min)	10.0a (9.0–12.0)	9.2 (8.0–11.0)	0.8	0.004
Total procedure time (min)	20.0b (16.0–26.5)	20.0 (15.0–27.6)	0.0	0.64

[95% CI] calculated using Wilson Score Interval for proportions where appropriate.

CADe, computer-aided detection; CI, confidence interval; IQR, interquartile range.

^aMissed data from 2 cases.

^bMissed data from 1 case. Data are n/N (%), n/N [95% CI], or median (IQR).

Table 3.

Additional analyses of primary and secondary outcomes in the modified intention-to-treat population

	CADe+ mucosal exposure device (N = 182)	CADe (N = 250)	Odds ratio/effect ratio CADe + MED to CADe	P value
Adenoma detection rate (ADR)a	86/182 (47.3%)	96/250 (38.4%)	1.16 [0.74–1.81]	0.51
Mean no. of adenomas per colonoscopy (APC)b	160/182 (0.88)	166/250 (0.66)	1.06 [0.79–1.42]	0.72
Sessile serrated lesion detection rate (SDR)c	23/182 (12.6%)	46/250 (18.4%)	0.67 [0.37–1.21]	0.19
Sessile serrated lesions per colonoscopy (SSLPC)d	28/182 (0.15)	76/250 (0.30)	0.75 [0.49–1.15]	0.19
Mean no. of polyps per colonoscopy (PPC)e	211/182 (1.16)	299/250 (1.20)	0.798 [0.62–1.02]	0.08

Data are n/N (%) [95% CI].

CADe, computer-aided detection; CI, confidence interval.

^aOdds ratio of a binary logistic regression mixed model analysis, with study site as clustering variable, and sex, index colonoscopy, colonoscopy indication, age, withdrawal time as confounding variables. Two cases were excluded due to missing data for withdrawal time.

^bEffect ratio of a negative binomial regression mixed model analysis, with study site as clustering variable, and sex, index colonoscopy, colonoscopy indication, age, withdrawal time as confounding variables. Two cases were excluded due to missing data for withdrawal time.

^cEffect ratio of a binary logistic regression mixed model analysis, with study site as clustering variable, and age and withdrawal time as confounding variables. 2 cases were excluded due to missing data for withdrawal time.

^dEffect ratio of negative binomial regression mixed model analysis, with study site as clustering variable, and sex, index colonoscopy, as confounding variables.

^eEffect ratio of a negative binomial regression mixed model analysis, with study site as clustering variable, and sex, index colonoscopy, colonoscopy indication, age, withdrawal time as confounding variables. Two cases were excluded due to missing data for withdrawal time.

An exploratory subgroup analysis was performed to compare ADR between groups for each of the 3 overlapping sites. At site 01, ADR was 40.7% in the CADe + MED group and 45.5% in the CADe group (P = 0.62; total colonoscopies with at least one adenoma: 22/54 vs 25/55). At Site 02, ADR was significantly higher in the CADe + MED group compared with CADe (56.1% vs 32.6%, P = 0.017; total colonoscopies with at least one adenoma: 32/57 vs 15/46). At Site 03, ADR was similar between groups (45.1% vs 44.4%, P = 0.94; total colonoscopies with at least one adenoma: 32/71 vs 24/54). Owing to the exploratory nature and no correction for multiple testing, results should be interpreted with caution. Additional outcomes are reported in Supplementary Digital Content (see Supplementary Table 2, 3 and 4, http://links.lww.com/CTG/B436, http://links.lww.com/CTG/B437, http://links.lww.com/CTG/B438).

DISCUSSION

In this multicenter study involving experienced endoscopists, we found that the addition of a MED to CADe-assisted colonoscopy in a diagnostic, non-FIT screening, and surveillance population was not associated with a statistically significant change in ADR compared with the CADe arm of a previously conducted RCT, after adjusting for clustering and confounders. Notably, both groups achieved ADRs well above established quality benchmarks (21,22), reflecting high procedural quality.

Although no significant association was observed, the ADR in our CADe + MED arm aligns with that of a recent multicenter RCT by Spadaccini et al (15), evaluating CADe with Endocuff Vision (Olympus, Tokyo, Japan), an attachable MED. This study included over 1,000 patients and reported a significantly higher ADR of 49.6% in the CADe + MED group vs 44.0% in the CADe-alone group, corresponding to a relative increase of 12%. Their study included FIT-referrals, whose ADR exceeded 60%; these patients were excluded in our study, which may have contributed to the higher overall ADR. In another RCT, Lui et al (16) also observed a significant increase in ADR (58.7% vs 53.8%) with CADe + Endocuff compared with CADe alone. However, in a recent RCT by Rocchetto et al (17) in a FIT-based screening program, the addition of a MED to CADe did not result in a significant increase in the detection of high risk colonoscopy lesions (19.3% vs 23.1%) and ADR (55.4% vs 59.1%), for CADe + MED vs CADe, respectively. Although both the G-EYE and Endocuff Vision are classified as MEDs, cross-trial comparisons should be interpreted with caution because of differences in design, deployment, and mucosal interaction.

We hypothesized a synergistic effect based on the distinct mechanisms of CADe and MED: CADe highlights already visualized lesions, whereas a balloon MED mechanically exposes mucosa that may otherwise remain concealed. However, the combined approach was not associated with a significant increase in ADR. This may partly reflect limited statistical power, given the smaller sample size compared with CADe + MED trials, with only a small absolute difference in ADR percentage points that may have been too small to detect in the current study sample (15,17). In addition, a ceiling effect may have contributed: among high-performing endoscopists, fewer additional adenomas may remain to be detected, limiting the potential incremental value of a MED. This hypothesis was also proposed by Rocchetto et al (17), who similarly found no significant benefit when combining CADe with a MED. Without tandem examination however the number of missed polyps cannot be approximated, and this potential ceiling effect therefore remains hypothesis-generating.

Placed in context, these findings align with the broader variability reported for CADe- and MED-assisted colonoscopy. Although studies and meta-analyses have largely shown improved detection with either approach (11,12,23,24), other studies found no benefit compared with conventional colonoscopy (2,18,25,26). This heterogeneity likely reflects differences in operator performance, case mix, and procedural factors influencing device effectiveness.

Besides the primary outcome, we observed variation in SSL detection and withdrawal time with the addition of a MED. SSL detection was lower in the CADe + MED group, although this difference was not significant after adjustment in the mixed-effects model. This contrasts with findings by Lui et al (16), who reported higher SSL detection with CADe + Endocuff Vision (42% vs 35%). However, sessile serrated lesion detection rate in that study was notably high, and it was not specified whether hyperplastic polyps were classified as SSLs. By contrast, both Spadaccini et al and Rocchetto et al found no improvement in SSL detection when a MED was added to CADe (15,17), with sessile serrated lesion detection rate in the Rocchetto study numerically lower in the combined arm (3.0% vs 5.7%, P = 0.11). Beyond limited power for SSL detection as a secondary outcome, differences in case-mix likely contributed: some high SSL-yield centers from the previous DISCOVERY II trial did not participate in this study, alongside the known inter-observer variability in SSL diagnosis (27).

Median withdrawal time was 10.0 minutes in the CADe + MED group compared with 9.2 minutes with CADe. Previous MED studies did not report an increased withdrawal time compared with conventional colonoscopy (2,11,15,28). In both groups, ADR increased progressively with longer withdrawal times, plateauing at approximately 10–12 minutes (see Supplementary Table 4, http://links.lww.com/CTG/B438). Although withdrawal time is a key quality indicator associated with higher ADR (21,29), the parallel increase in both groups suggests it acted as a general quality driver rather than explaining the absence of a group difference. Importantly, this factor was accounted for in the mixed-effects model.

In a broader context, although an increase in ADR in conventional colonoscopy is inversely associated with the risk of CRC (30), the effect of an increased ADR with CADe, MED, or their combination on long-term outcomes such as postcolonoscopy CRC incidence and mortality has yet to be studied. Although cost-effectiveness models suggest potential benefit (31), society guidelines remain cautious. Recently, the European Society of Gastrointestinal Endoscopy issued a position statement with a weak recommendation for the use of CADe, whereas the American Gastroenterological Association concluded that as yet no recommendation can be made. Both societies emphasized as main reason the insufficient evidence on clinically meaningful long-term outcomes (32,33).

The strengths of this study include the participation of experienced endoscopists already proficient with CADe and inclusion criteria that mirror a non-FIT screening clinical practice. This study also has limitations. First, the nonrandomized design and use of historical controls introduced a potential risk of selection and time-related bias, despite adjustments using mixed-effects regression. Second, although the 3 highest-enrolling sites from the previous CADe study were included, 4 CADe-only sites were not, potentially affecting case mix. However, more than 60% of the procedures in both groups were performed at similar sites. Third, procedures were conducted in partially overlapping but nonidentical timeframes, making the study susceptible to temporal variation in practice. Still, most procedures in this study were performed by the same endoscopists as in our comparative cohort, making consistency in endoscopy practice over time likely. Fourth, we cannot rule out the possibility that mechanical deformation from the MED altered the visual field in a way that could interfere with CADe performance.

In conclusion, in a diagnostic, non-FIT screening, and surveillance colonoscopy, CADe combined with a MED did not significantly improve ADR over CADe alone, suggesting limited synergy in high-performing centers.

CONFLICTS OF INTEREST

Guarantor of the article: Michiel H.J. Maas, MD.

Specific author contributions: M.H.J.M.: conceptualization; formal analysis; project administration; validation; visualization; writing–original draft; writing–review & editing; M.L.M.v.R.: conceptualization; formal analysis; project administration; validation; visualization; writing–original draft; writing–review & editing; T.R.: investigation; methodology; resources; writing–review & editing; D.S.: investigation; project administration; writing–review & editing; P.D.S.: conceptualization; funding acquisition; investigation; methodology; project administration; resources; supervision; writing–review and editing.

Financial support: This study was supported by a research grant provided by the ESGE Research Champions' Den and an unrestricted grant from PENTAX Medical, Tokyo, Japan.

Potential competing interests: M.H.J.M., M.L.M.v.R., M.L.M.v.R.: none; T.R.: received speaker fees from: Olympus Medical, Pentax Medical, Mauna Kea Technologies, Medtronic, Takeda, Galapagos, Falk, Janssen, Abbvie, Repha, Medical Tribune, Lilly; P.S.: none; D.S.: none; P.D.S.: research support from MicroTech, Magentiq Eye, VTM Technologies, Norgine and Pentax.

IRB statement: This study was approved by independent institutional review boards at each site and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. The study was registered and approved under identifier 2021-13357 at the site of the sponsor.

Study Highlights.

WHAT IS KNOWN

• ✓ Computer-aided detection (CADe) highlights visible adenomas during colonoscopy.

• ✓ Mucosal exposure devices (MEDs) improve mucosal visualization by flattening colonic folds.

• ✓ Both CADe and MED independently increase adenoma detection rates (ADR).

WHAT IS NEW HERE

• ✓ Adding a MED to CADe-assisted colonoscopy was not independently associated with increased ADR.

• ✓ Limited incremental benefit may suggest a ceiling effect in high-performing centers.

Supplementary Material

ct9-17-e00963-s001.docx ^{(21KB, docx)}

ct9-17-e00963-s002.docx ^{(15.9KB, docx)}

ct9-17-e00963-s003.docx ^{(17.2KB, docx)}

ct9-17-e00963-s004.docx ^{(16KB, docx)}

ct9-17-e00963-s005.docx ^{(14.9KB, docx)}

ABBREVIATIONS:

ADR

adenoma detection rate

APC

adenomas per colonoscopy

BBPS

Boston Bowel Preparation Score

BMI

body mass index

CADe

computer-aided detection

CI

confidence interval

CONSORT-AI

consolidated standards of reporting trials - artificial intelligence

CRC

colorectal cancer

ESGE

European Society of Gastrointestinal Endoscopy

FIT

fecal immunochemical test

GCP

good Clinical Practice

IRB

Institutional Review Board

IQR

interquartile range

MED

mucosal exposure device

PDR

polyp detection rate

PPC

polyps per colonoscopy

RCT

randomized controlled trial

SD

standard deviation

SDR

sessile serrated lesion detection rate

SSL

sessile serrated lesion

SSLPC

sessile serrated lesions per colonoscopy