From availability to adoption: Why endoscopist-level use determines the real-world impact of computer-aided detection in colonoscopy

Abstract

Colonoscopy quality, commonly measured by adenoma detection rate (ADR), remains a cornerstone of colorectal cancer prevention. While computer-aided detection (CADe) systems have demonstrated efficacy in controlled trials, their real-world impact has been inconsistent, raising questions about how these technologies should be implemented in daily practice. The recent real-world observational study by Rao et al at an Australian tertiary center provides important insight by shifting attention from technology availability to endoscopist-level adoption. By stratifying endoscopists according to CADe usage frequency, the study demonstrates that even partial use of CADe is associated with meaningful improvements in ADR, while high-frequency use confers additional gains in adenoma yield per procedure. These findings suggest that CADe effectiveness is not an intrinsic property of the technology alone but is largely influenced by user behavior and integration into routine endoscopic workflows. From a gastroenterology perspective, this has important implications: Inconsistent or selective use of CADe may blunt its potential benefits, whereas deliberate adoption strategies could enhance polyp detection and risk stratification. Future efforts should therefore move beyond simple deployment toward optimizing training, adherence, and quality frameworks that encourage appropriate and sustained CADe use in clinical practice. Importantly, this editorial advances prior technology-centered discussions by explicitly framing endoscopist-level adoption intensity as a central explanatory factor for heterogeneous real-world CADe outcomes. From a practice and policy perspective, this adoption-oriented framework highlights the need for structured monitoring, training, and integration of CADe into quality assurance systems rather than passive technological deployment.

Key Words: Computer-aided detection; Colonoscopy; Adenoma detection rate; Endoscopist behavior; Human-artificial intelligence interaction; Real-world implementation

Core Tip: Computer-aided detection (CADe) systems have proven efficacy in improving adenoma detection rate during colonoscopy, yet their real-world impact remains heterogeneous. Emerging evidence suggests that CADe effectiveness is not determined by mere availability but by endoscopist-level adoption and integration into routine practice. Stratifying endoscopists by CADe usage frequency reveals a dose-response relationship between use intensity and adenoma yield. These findings underscore the need to shift implementation strategies from passive deployment toward active training, adherence, and behavioral optimization to fully realize the clinical value of CADe, highlighting endoscopist-level adoption as a key translational determinant beyond mere technological availability.

This editorial refers to “Availability and use of computer-aided detection during colonoscopy: A real-world observational study at an Australian tertiary center” by Rao et al, 2026; https://doi.org/10.3748/wjg.v32.i4.112698.

---

INTRODUCTION

Colonoscopy remains the cornerstone of colorectal cancer (CRC) prevention, with adenoma detection rate (ADR) serving as a validated surrogate for post-colonoscopy CRC risk[1,2]. Despite advances in endoscopic imaging and quality metrics, ADR continues to exhibit substantial inter-operator variability, even among experienced endoscopists[3]. This persistent variability has driven the development and clinical adoption of artificial intelligence (AI)-based computer-aided detection (CADe) systems aimed at standardizing mucosal inspection and reducing lesion miss rates.

Randomized controlled trials and multicenter studies have consistently demonstrated that CADe improves ADR and adenomas per colonoscopy (APC), particularly for diminutive and flat lesions[4-6]. These data have supported regulatory approval and rapid diffusion of CADe into clinical practice. However, real-world observational studies have yielded conflicting results, with some reporting modest benefit and others showing no improvement-or even paradoxical declines-in detection outcomes following CADe implementation[7,8].

This discrepancy highlights a critical but underappreciated factor: The endoscopist, not the algorithm alone, may substantially influence CADe effectiveness. Recent real-world data suggest that variability in CADe usage frequency, trust, and workflow integration at the individual endoscopist level may largely explain heterogeneous outcomes[9]. Notably, a recent observational study by Rao et al[10] conducted at a tertiary center stratified endoscopists according to CADe usage frequency and reported that even partial use was associated with improved ADR, while higher adoption intensity yielded additional gains in APC, thereby supporting a graded adoption-outcome relationship in routine clinical practice.

This editorial critically interprets current evidence to explore why endoscopist-level use is closely associated with the real-world impact of CADe in colonoscopy, emphasizing behavioral adoption, human-AI interaction, and procedural context as key determinants of clinical benefit (Figure 1 and Table 1). Figure 1 and Table 1 are intended as conceptual synthesis frameworks to enhance interpretability of heterogeneous evidence rather than exhaustive evidence-mapping summaries. Unlike prior editorials that primarily focus on algorithmic performance or technological availability, the present perspective explicitly proposes an adoption-centered interpretative framework and discusses its direct implications for real-world implementation, quality monitoring, and policy-oriented deployment of CADe in routine colonoscopy practice. However, these associations should be interpreted with caution, as observational real-world studies may be influenced by confounding factors such as baseline endoscopist skill, case-mix, and selective CADe activation. Key potential confounders include baseline adenoma detection performance, withdrawal time, bowel preparation quality, and endoscopist experience level; although the referenced real-world study partially addressed these factors through stratification and adjusted analyses, residual confounding and selection bias cannot be fully excluded.

Figure 1 This conceptual framework illustrates how the real-world effectiveness of computer-aided detection in colonoscopy is influenced by endoscopist-level adoption rather than technology availability alone, based on an integrated synthesis of evidence from clinical trials, observational studies, and other research. CADe: Computer-aided detection.

Table 1 Integrated determinants of real-world effectiveness of computer-aided detection in colonoscopy: An evidence-informed conceptual synthesis.

Domain	Key determinant	Integrated evidence synthesis	Impact on CADe effectiveness
Technology	CADe availability	The mere presence of an approved CADe system enables real-time polyp alerts but does not ensure improved detection outcomes without active user engagement	Necessary prerequisite but insufficient alone
	Alert characteristics	Systems with excessive false-positive alerts increase cognitive load and distract visual attention, whereas low false-positive systems facilitate trust and sustained use	Modulates trust and attention
Endoscopist behavior	Usage frequency	Consistent or high-frequency CADe use is associated with incremental improvements in adenoma detection, while intermittent or selective use yields attenuated benefit	Establishes a dose-response relationship
	Adoption intensity	Adoption intensity, rather than binary use, determines the magnitude of adenoma yield and detection consistency	Primary determinant of real-world impact
Endoscopist expertise	Baseline detection skill	CADe disproportionately benefits trainees and low baseline detectors, narrowing performance gaps with experts; benefits among high detectors are smaller and context-dependent	Differential effect by skill level
	Ceiling effect	In high-performing endoscopists, CADe gains may plateau or be limited to subtle lesions, reflecting a ceiling effect rather than lack of efficacy	Explains heterogeneous results
Human–AI interaction	Trust calibration	Balanced trust enhances responsiveness to meaningful alerts, whereas overreliance or skepticism reduces CADe utility	Determines effective alert utilization
	Cognitive load	Excessive alerts or workflow disruption increase mental burden and may impair visual search strategies	Can blunt or reverse benefit
Workflow integration	Procedural compatibility	Seamless integration into routine withdrawal and inspection patterns facilitates adoption, while disruptive interfaces reduce sustained use	Influences adherence
	Response behavior	Effective CADe use depends on how endoscopists respond to alerts, not merely alert presence	Mediates clinical translation
Procedural quality	Withdrawal time	Adequate withdrawal time synergizes with CADe to enhance detection, while rushed examinations limit AI benefit	Enables CADe amplification
	Bowel preparation	High-quality bowel preparation maximizes CADe sensitivity and downstream detection outcomes	Foundational modifier
Clinical outcomes	ADR	ADR improvement is most pronounced with consistent CADe adoption and among low-to-intermediate baseline detectors	Primary quality endpoint
	Adenomas per colonoscopy	High adoption intensity is associated with additional gains in adenoma yield beyond ADR alone	Reflects cumulative benefit
	Miss rate variability	CADe adoption reduces inter-endoscopist variability and fatigue-related declines when consistently applied	Improves quality stability
Implementation strategy	Passive deployment	Unstructured implementation without training or monitoring results in inconsistent use and muted benefit	Limits real-world effectiveness
Active adoption framework	Training, adherence monitoring, and workflow optimization promote sustained CADe use and maximize clinical value	Optimizes translation to practice

The determinants summarized in this table represent an integrated conceptual synthesis derived from heterogeneous evidence sources, including randomized controlled trials, real-world observational studies, and other literature, rather than a claim-by-claim systematic evidence map. ADR: Adenoma detection rate; AI: Artificial intelligence; CADe: Computer-aided detection.

ENDOSCOPIST-LEVEL ADOPTION AS THE PRIMARY DETERMINANT OF CADE EFFECTIVENESS

Early CADe trials implicitly assumed uniform usage across operators; however, real-world studies reveal marked heterogeneity in how often and how consistently endoscopists activate CADe during procedures. In the real-world cohort referenced above, endoscopists were categorized by low, partial, and high CADe utilization, and a stepwise increase in ADR and APC was observed across these adoption strata, further reinforcing the behavioral dose-response hypothesis[10,11]. Stratification of endoscopists by CADe usage frequency demonstrates a graded association with ADR improvement, suggesting a behavioral dose-response effect rather than an all-or-none technological phenomenon[12,13].

Importantly, even partial or intermittent CADe use has been associated with meaningful gains in ADR compared with no use, while high-frequency use confers additional increases in APC and detection of subtle lesions[14,15]. This finding challenges the assumption that CADe benefit requires universal or continuous deployment and instead underscores the role of individual engagement and intentional use[16]. Nevertheless, interpretation of these real-world findings should consider potential selection effects, as high-frequency CADe users may also represent operators with greater baseline quality awareness and procedural diligence[17].

These observations align with broader quality improvement literature showing that performance-enhancing technologies exert maximal effect only when actively embraced by operators. CADe should therefore be conceptualized not as an autonomous diagnostic tool but as a cognitive adjunct whose value emerges through consistent human interaction.

DIFFERENTIAL IMPACT ACROSS ENDOSCOPIST EXPERIENCE LEVELS

A consistent theme across studies is that CADe disproportionately benefits endoscopists with lower baseline ADRs, including trainees and community practitioners, effectively narrowing performance gaps with experts[18-20]. In these groups, CADe functions as a visual attention scaffold, reducing miss rates and enhancing confidence during mucosal inspection.

Conversely, among high-performing or expert endoscopists, CADe effects are more variable. Several studies report modest gains limited to small or flat adenomas, while others show no improvement or even reduced detection, potentially reflecting ceiling effects or altered visual search strategies[8,21].

These findings suggest that CADe does not uniformly amplify performance but interacts dynamically with baseline skill, visual expertise, and cognitive workload. As such, endoscopist-level adoption strategies may need to be tailored according to experience and baseline detection performance rather than applied uniformly.

HUMAN–AI INTERACTION: TRUST, COGNITIVE LOAD, AND BEHAVIORAL INTEGRATION

Beyond frequency of use, the quality of human-AI interaction critically shapes CADe effectiveness. Qualitative and eye-tracking studies demonstrate that CADe modifies endoscopists’ visual attention patterns, sometimes acting as a “second observer” but, in other cases, inducing overreliance or alert fatigue[22,23].

False-positive alert burden represents a key determinant of trust and sustained use. Systems with higher false-positive rates increase cognitive load, prolong procedure time, and may paradoxically reduce detection accuracy by distracting from true lesions[24,25]. In contrast, CADe platforms optimized for specificity appear to foster greater endoscopist confidence and more effective workflow integration.

These data emphasize that CADe adoption is not merely a technical issue but a behavioral one, influenced by usability, interpretability, and endoscopist digital literacy[26,27]. In parallel, emerging imaging-enhanced AI approaches, such as hyperspectral and spectral imaging–assisted CADe/CADx systems in esophageal neoplasia, further illustrate how advances in lesion visualization and classification may interact with operator adoption and workflow integration, although their real-world effectiveness will likewise depend on clinician-level implementation rather than technological capability alone[28,29]. Without targeted training and feedback, inconsistent use or selective disengagement may blunt CADe’s real-world impact.

PROCEDURAL CONTEXT AND WORKFLOW INTEGRATION

CADe performance remains tightly coupled to fundamental procedural quality metrics. Adequate withdrawal time and bowel preparation consistently amplify CADe benefit, whereas suboptimal technique limits detection regardless of AI assistance[3,30].

Notably, CADe appears to mitigate fatigue-related declines in ADR during late-day procedures, suggesting a role in sustaining vigilance over prolonged endoscopy lists[31]. However, this benefit is contingent on active engagement rather than passive availability.

These findings reinforce that CADe cannot compensate for poor technique but may enhance high-quality practice when embedded within established quality frameworks.

IMPLICATIONS FOR IMPLEMENTATION AND QUALITY IMPROVEMENT

Collectively, current evidence supports a paradigm shift in CADe implementation at the unit and institutional level-from technology-centered deployment to endoscopist-centered adoption strategies. Simply installing CADe systems without addressing training, adherence, behavioral integration, and structured quality monitoring risks underutilization and disappointing real-world outcomes.

Future implementation models should incorporate: (1) Endoscopist-level monitoring of CADe usage frequency; (2) Targeted training focused on trust calibration and false-positive management; and (3) Integration with quality metrics such as withdrawal time and APC.

Feedback mechanisms that reinforce sustained and appropriate use. In practical terms, implementation may include routine tracking of the proportion of colonoscopies performed with active CADe, stratification of adoption intensity (e.g., low, partial, and high use), and periodic audit-and-feedback cycles integrated into existing endoscopy quality assurance programs.

Such strategies may maximize CADe’s potential to improve detection while minimizing unintended consequences such as overreliance or deskilling.

CONCLUSION

The real-world impact of CADe in colonoscopy is influenced less by algorithmic capability than by how endoscopists choose to use it. Evidence increasingly demonstrates that endoscopist-level adoption-encompassing usage frequency, trust, and workflow integration-is a major contributor to variability in CADe effectiveness. Partial use yields measurable benefit, while consistent engagement confers additional gains in adenoma yield.

Recognizing CADe as a human-dependent technology reframes implementation priorities toward training, behavioral optimization, and quality integration. Future efforts to improve colonoscopy outcomes should therefore focus not only on improving algorithms but on enabling endoscopists to adopt CADe deliberately, consistently, and effectively in everyday clinical practice.