Early detection and resection of colorectal polyps remain the cornerstones of colorectal cancer prevention. The well‐known adenoma‐carcinoma cascade involved in colorectal cancer development presents a crucial opportunity for early intervention before more advanced cancers develop [1]. Despite decades of progress in endoscopic imaging, interval colorectal cancers still occur—often the result of elusive precancerous lesions going undetected at colonoscopy [2]. Increasingly, the culprit lesions are subtle: sessile serrated lesions (SSLs), flat polyps, and those hidden in the more proximal colon [2, 3]. These account for a substantial share of missed prevention opportunities and have driven the push for more sophisticated imaging technology and adjunct devices.

Texture and Color Enhancement Imaging (TXI) is one such innovation, designed to amplify brightness, contrast, and mucosal texture, making subtle abnormalities more conspicuous [4]. In a 2024 randomized controlled trial including 324 patients, we demonstrated that the use of TXI improved the adenoma detection rate (ADR) from 40.99% with white light imaging (WLI) to 54.6% (p = 0.01), nearly doubling the mean number of adenomas per colonoscopy (APC) (1.71 with TXI vs. 0.94 with WLI, p < 0.01) [5]. Notably, enhanced detection was most pronounced for flat and right‐sided lesions, as has been suggested in prior visibility studies [5, 6]. Although a 2025 RCT subsequently found no difference in adenoma detection with TXI (excluding SSLs), the effect remained significant for subtle flat polyps, such as SSLs [7]. These studies have raised an important consideration: whether there is still the opportunity to further enhance adenoma detection when detection rates are already this high.

Artificial intelligence–assisted polyp detection systems, labeled “computer‐aided detection (CADe),” have emerged concurrently as a powerful adjunct for endoscopic detection of gastrointestinal lesions [8]. The impact of CADe systems during colonoscopy was illustrated by a recent meta‐analysis including 23 861 patients, which demonstrated a 20% increase in ADR (risk ratio 1.2, p < 0.01) using CADe versus routine unassisted colonoscopy [9]. However, CADe systems are associated with high rates of false positives, leading to proceduralist fatigue and potentially prolonging withdrawal times [10].

Theoretically, pairing TXI's enhanced visualization with CADe's vigilance should produce a synergistic effect—two complementary technologies tackling optimal polyp detection from different angles. However, as polyp miss rates decline, the gains from additional adjuncts may diminish. This study investigates whether the combination of CADe with TXI can deliver additional gains in ADR substantial enough to justify its potential drawbacks.

A total of 263 patients were included in the final analysis—100 patients in the TXI and CADe group and 163 patients in the historical control group (Table 1). A further four patients were consented for inclusion in the intervention arm but excluded due to inadequate bowel preparation. There were no differences in demographics or the indications for colonoscopy between the two groups (Table 1). The number of procedures performed by each of the involved proceduralists was appropriately matched (p > 0.99). There was a significant increase in mean BBPS in the TXI and CADe group at 8.31 versus 7.91 (p = 0.014).

There was no difference in the primary outcome of the number of APC, at 1.71 in both groups (p = 0.99). Similarly, there was no difference in the overall ADR (62% in the intervention arm vs. 54.6% in the control arm, p = 0.33). There was also no difference in the number of advanced polyps between the two groups (0.24 in the TXI and CADe arm and 0.43 in the TXI‐only arm, p = 0.28). There was no difference in the detection of flat (0.6 vs. 0.58, p = 0.89) or right‐sided (1.19 vs. 1.13, p = 0.79) polyps. There was a numerical but nonsignificant trend toward increased SSL detection using the combination of TXI and CADe at 0.31 SSLs per colonoscopy versus 0.17 in the TXI‐only group (p = 0.17), with an SSL detection rate of 14% versus 10.4% (p = 0.38).

There was a significant increase in withdrawal time using TXI and CADe, at a mean of 8.15 min in the intervention arm compared to 7.17 in the control arm (p < 0.001) (Table 2). There was also a significant increase in resection of non‐neoplastic polyps at 27% in the TXI and CADe group compared to 16.6% in the TXI‐only group (p = 0.042). A subgroup analysis limited to patients with indications according to the NBCSP demonstrated no difference in ADR in the intervention versus control groups (73.8% vs. 64.3%, p = 0.20) and no difference in APC (2.26 vs. 2.22, p = 0.94).

A multivariable negative binomial regression for the number of adenomas detected demonstrated no increase in adenoma detection with TXI and CADe (IRR = 1.0, p = 0.99) (Table 3). There was a trend toward an increased number of adenomas in males (aIRR = 1.37, p = 0.065), whereas longer withdrawal time was associated with increased adenoma detection (aIRR = 1.01, p < 0.001), and indications of iron deficiency anemia (aIRR = 0.47, p = 0.032) and altered bowel habit (aIRR = 0.25, p = 0.003) held negative associations.

When stratified according to proceduralists' overall ADR (with a cut‐off for “high detectors” being ADR > 55%), there was no difference in ADR with TXI and CADe compared to TXI alone in high detectors (62% vs. 66%, p = 0.57). In standard detectors, there was a significant improvement in ADR in the intervention group (63% vs. 44%, p = 0.046). Similarly, whereas there was a significant difference in ADR between high detectors and standard detectors (66% vs. 44%, p = 0.005) using TXI alone, this difference was no longer present when TXI was combined with CADe (63% vs. 62%, p > 0.9).

This study has demonstrated no difference in overall adenoma detection when combining CADe with TXI compared to TXI alone. There was no difference in right‐sided or flat polyp detection—lesions that are notoriously difficult to identify endoscopically and therefore key targets for CADe. As such, it appears that when high‐quality colonoscopy using TXI is performed by experienced proceduralists with high baseline ADR, a plateau for adenoma detection may have been reached, leaving no incremental gain with the use of CADe.

Although there was no impact on adenoma detection overall, there was a significant improvement in ADR when CADe was combined with TXI for proceduralists whose overall ADR was < 55%. A threshold of 55% was chosen not only based on our previous study comparing TXI with WLI but also based on a large 2022 audit, which determined this threshold to be associated with high detection of subtle polyps such as proximal serrated lesions [5, 12]. In our study, the significant difference in ADR between high detectors and standard detectors with TXI alone was no longer seen in the CADe group. This divergence has been demonstrated in prior studies, with a 2022 tandem colonoscopy study reporting that trainee endoscopists with CADe achieved equivalent ADR and APC to experienced endoscopists [13, 14]. Similarly, a recent study from Djinbachian et al. stratified the effect of AI according to varied withdrawal times and found no impact of CADe on ADR in procedures where the withdrawal time was > 8 min but a significant improvement in ADR when the withdrawal time was < 6 min [15]. This supports the assertion that the main benefit of CADe may be in ensuring a universal standard of care is reached rather than further improving the already high ADRs of expert interventionalists.

The impact of CADe on withdrawal times during colonoscopy has been a point of controversy. Whereas multiple large meta‐analyses have demonstrated prolonged withdrawal times with CADe, subgroup analyses including only patients without polyps have demonstrated no significant difference [9, 16, 17, 18, 19, 20]. It has therefore been postulated that the impact of CADe on withdrawal time is primarily attributable to improved polyp detection. Hassan et al. performed a post hoc analysis of a randomized trial comparing CADe with WLI, evaluating videos of colonoscope withdrawal using CADe [10]. In their study, false positives resulted in a negligible 1% increase in total withdrawal time, as the majority were immediately discarded by the endoscopist. Nevertheless, this was a video‐based study examined by two expert endoscopists with experience in AI. In the real world, where the majority of endoscopists do not have expertise in endoscopic imaging and factors such as bowel preparation can be highly variable, the impact of these false positives would be expected to be far more significant. This effect was pronounced in our study, with an almost 1‐min, or 13.7%, increase in withdrawal time using CADe versus TXI alone. The additional detail revealed by TXI may have amplified the impact on withdrawal time by enhancing subtle changes within these false positives, thereby requiring additional inspection time to exclude flat polyps. In addition, as the CADe system is trained by WLI images, not TXI, using TXI as the input may have increased the detection of false positives.

Another important consideration when implementing CADe on a widespread scale is the potential to increase unnecessary polypectomies. In our study, there was a significant increase in the resection of non‐neoplastic polyps (27% using TXI and CADe vs. 16.56% with TXI alone, p = 0.042). Recent studies, including a large meta‐analysis, have similarly demonstrated an increase in non‐neoplastic polyp resection [9, 21]. However, these are generally diminutive polyps resected with a cold snare and are therefore associated with negligible risk of complication. Nevertheless, clinical trials in CADe have predominantly been performed at expert centers with experienced proceduralists capable of accurately assessing false positives and non‐neoplastic polyps. One concern in our cohort was the frequent misidentification of inverted diverticula or even the appendiceal orifice by the CADe system (Figure 1). Inexperienced operators who are more likely to be influenced by the CADe system may attempt snare resection of these “lesions,” risking perforation or appendicitis.

Interestingly, there was a statistically significant difference in BBPS between the intervention and control groups in our study, favoring the CADe group. There were no procedural or logistical changes at the involved sites in terms of pre‐procedure preparation to account for this discrepancy. It is possible that the difference in bowel preparation was related to either chance or unmeasured differences between the two study periods. However, we would postulate that this difference may reflect a “false sense of security” related to the use of CADe, leading to bias toward overestimating the adequacy of mucosal views when CADe is used. This may have been reflected in higher BBPS—a subjective score documented by the endoscopist at the time of the procedure. Similar concerns have been raised in real‐world implementation studies with CADe, where the dramatic improvements seen in randomized controlled trials have not been reflected. In their 2023 study, Ladabaum et al. found no effect on ADR, APC, or any other detection metric, postulating that the difference between their study and previous randomized controlled trials implies unmeasured effects on endoscopist behavior when exposed to AI [22]. Levy et al. found that implementation of CADe resulted in a reduction in ADR and shorter mean procedural time (including for the subgroup without polyps) in their retrospective observational study—likely related to the impact of AI on human behavior [23]. Additionally, a recent study by Budzyń et al. has highlighted the issue of proceduralist “deskilling” after exposure to CADe‐assisted colonoscopy, with a 6% reduction in ADR [24]. Although our study did not explicitly examine the longitudinal effect of CADe on endoscopist behavior, the over‐reliance on AI and “false sense of security” are key concerns that should be carefully considered before the implementation of these systems becomes more widespread.

This is the first study evaluating the additive benefit of CADe to advanced imaging techniques such as TXI for polyp detection. It highlights important considerations with the implementation of CADe—the potential that the limit of adenoma detection is already reached when judicious colonoscopy is performed by highly trained proceduralists using image‐enhancement technology, as well as potential pitfalls of prolonged withdrawal time and unnecessary polypectomy, albeit acknowledging the potential for CADe to ensure a universal standard of care is maintained. However, there are limitations. Firstly, this is not a randomized controlled trial but instead uses historical controls and therefore may introduce bias related to unmeasured confounders. Nevertheless, the study was performed under the exact same conditions as the historical control group, with all patients in the intervention arm prospectively recruited. According to our demographic data, there were no significant differences between the groups, supporting the validity of our methods. A further limitation is that the effect size of AI estimated for our power calculation was not borne out in our study, and therefore, the study may have been underpowered. However, if the effect size for CADe is in fact smaller than this, its additive benefit is unlikely to be of clinical significance and may not outweigh the potential drawbacks. Our study was also not adequately powered to confidently assess the impact of CADe combined with TXI on lesion subtypes such as SSL, flat, and right‐sided polyps. Finally, as mentioned above, the CADe training data set did not include TXI images. Although this may have influenced its reliability, this is considered less likely as TXI involves digital, not optical, enhancement technology.

In summary, the addition of CADe did not improve adenoma detection beyond that of TXI alone. However, whereas high‐detecting interventional endoscopists may not benefit, CADe may still have a role in ensuring a universal standard of care—maintaining an additive benefit for many endoscopists. Before the widespread implementation of CADe, concerns need to be addressed regarding non‐neoplastic polyp resection, increased withdrawal times, and proceduralist deskilling. In the meantime, the use of image enhancement technology such as TXI, combined with careful mucosal examination by appropriately skilled endoscopists, will result in highly effective colonoscopic screening and surveillance.

The authors received no funding for this study.

This study was approved by the Central Adelaide Local Health Network Human Research Ethics Committee according to the National Health and Medical Research Council National Statement on Ethical Conduct in Human Research (2007). Informed consent was obtained from all patients prior to participation.

The authors declare no conflicts of interest.

We acknowledge Aline Willsmore for her assistance with the statistical analysis. Open access publishing facilitated by Adelaide University, as part of the Wiley ‐ Adelaide University agreement via the Council of Australasian University Librarians

The data that support the findings of this study are available from the corresponding author upon reasonable request.