Teaching Six ‘Blink’ Features Reduces General Endoscopists’ Cancer Miss-Rate in Image-Based Assessment of Large Colorectal Polyps

Teaching Six ‘Blink’ Features Reduces General Endoscopists’ Cancer Miss-Rate in Image-Based Assessment of Large Colorectal Polyps | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Teaching Six ‘Blink’ Features Reduces General Endoscopists’ Cancer Miss-Rate in Image-Based Assessment of Large Colorectal Polyps Lynn K. Debels, Tamas Tornai, John Anderson, Maria Eva Argenziano, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8553404/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background & Aims: Accurate cancer detection in large non-pedunculated colorectal polyps (LNPCPs) remains challenging for general endoscopists. We evaluated whether teaching six gross morphological "Blink" features could improve the accuracy of cancer detection. Methods This prospective interventional study assessed general endoscopists evaluating 20 LNPCP images (7 with histologically-confirmed submucosal invasive cancer including 4 deep invasions ≥ 1000µm, 13 benign). Participants assessed images before and after a 2-minute educational video introducing six Blink features: spontaneous bleeding, depression, fold deformation, extra redness, ulceration, and chicken-skin mucosa. Primary outcome was change in miss-rate for cancer detection. Generalized linear mixed models accounted for clustering within raters and polyps. Results 165 participants included gastroenterology consultants (63.6%), trainees (21.2%), students (1.8%) and colorectal surgeons (13.3%) with a median colonoscopy experience of 6.5 years. Post-intervention the cancer miss-rate decreased four-fold from 26.6% (95%CI 13.4–46.0) to 5.7% (95%CI 2.4–12.8). This improvement was consistent across experience levels. The false alarm rate increased less-than two-fold from 25.0% (95%CI 15.1–38.5) to 42.2% (95%CI 27.7–58.2). Multivariable analysis identified spontaneous bleeding (OR 3.92, 95%CI 3.11–4.96), extra redness (OR 3.66, 95%CI 3.09–4.33), and depression (OR 3.06, 95%CI 2.58–3.64) as independent predictors of cancer amongst general endoscopists. Mean Blink features per polyp were 1.08 (95%CI 0.82–1.40) for benign lesions versus 2.46 (95%CI 1.85–3.12) for cancers. Conclusions Teaching six Blink features to general endoscopists led to a 4x reduction in cancer miss-rates in an image-based evaluation. While specificity decreased, this trade-off favors patient safety, as false-positives trigger established clinical safeguards while missed cancers risk inappropriate endoscopic resection with potentially irreversible consequences. Graphical Abstract Figures Figure 1 Figure 2 Figure 3 Introduction Colorectal cancer is one of the most preventable malignancies through the early detection and removal of precancerous lesions via polypectomy[ 1 , 2 ]. The technique required to endoscopically resect a given polyp is related to the presence of cancer within that lesion and the estimated depth of any submucosal invasion (SMI)[ 3 ]. Incorrect decision making can lead to morbidity and potential mortality for patients and costs for healthcare systems[ 4 ]. Real-time polyp assessment for cancer risk using high-resolution endoscopy with magnification and virtual chromoendoscopy is effective amongst experts[ 5 , 6 ], but remains complex for general endoscopists without specific training[ 7 , 8 ]. Since these practitioners will perform the majority of procedures where large (≥ 10mm) non-pedunculated colorectal polyps (LNPCPs) are first detected there is a pressing need for simplified and robust approaches to allow estimation of malignant potential and effective triage of treatment strategy. It is well recognized that human experts, regardless of the domain of expertise, are exceptionally adept at synthesizing large volumes of visual, contextual, and historical information to arrive at a first (or “Blink”) impression[ 9 ]. In medicine there is precedent for the Blink concept. One study[ 10 ] demonstrated that expert radiologists who received only brief exposure to mammograms could detect breast cancer as accurately as when they were allowed unlimited exposure to the same mammograms. In another study[ 11 ], both radiologists and cytologists were able to reliably stratify medical images as normal/abnormal during brief exposures (250–2000 milliseconds) without being able to accurately localise the actual pathology during that exposure. Importantly both studies demonstrated that non-experts could not reliably detect the outcome based on brief exposures. Applied to LNPCPs, the success of expert Blink impression likely hinges on the ability to quickly recognize specific visual features that are strongly associated with underlying malignancy when first encountering the polyp. It follows that these cues would be morphologic features visible from afar and without virtual chromoendoscopy, since there would be no time for detailed analysis of pit/vascular pattern within the required timeframe. So-called gross morphologic features, shown to significantly increase the risk of a polyp containing deep submucosally invasive cancer when identified by experts[ 12 ], would be excellent candidates. The authors proposed a condensed set of gross morphological characteristics which could potentially be recognized at-a-glance by general endoscopists, mimicking the intuitive pattern matching used by experts. It was hypothesized that such an approach could shortcut the need for significant experience to identify cancer in LNPCPs. The six selected features, hereafter referred to as “Blink features”, were spontaneous bleeding, depression, fold deformation, extra redness, ulceration and chicken-skin mucosa [ 13 – 28 ] (Fig. 1 , Suppl. Materials – Narrative Review ). This study aimed to assess the impact of introducing a structured assessment of the six Blink features using a 2-minute learning intervention on the diagnostic performance of cancer prediction from still images of LNPCPs amongst general endoscopists. Methods Study Design This was a prospective interventional study designed to evaluate the accuracy of colorectal cancer detection using Blink impression before and after the introduction of six Blink features using a 2-minute online learning intervention. The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of Ghent University Hospital (ONZ-2022-0488). The study was registered on ClinicalTrials.gov (NCT05699954) on 20 April 2023. Informed consent was obtained from participating patients and endoscopists prior to the study. Phase 1 – Selection of Blink Features A comprehensive literature review was conducted in Medline, Embase, PubMed, and the Cochrane Library to identify gross morphological features of colorectal polyps associated with deep SMI ( Supplementary Materials , Suppl. Table 1 ). From the identified literature, six features were selected by author consensus (DJT, JA, LDs, LDb) based on three criteria: (1) strength of evidence linking the feature to deep SMI, (2) clinical relevance and frequency of observation, and (3) ease of identification by endoscopists of varying experience using white-light endoscopy without magnification or image enhancement. The selected features, collectively referred to as the six Blink features were: 1. Fold Deformation – convergence of ≥ 3 adjacent haustral folds towards, or interruption of an existing fold by, the lesion. 2. Extra Redness – one or more foci of mucosal erythema on the lesion surface that are a deeper hue than the immediately-adjacent polyp tissue . 3. Depression – a clearly demarcated concavity or excavated area on the luminal surface of the polyp . 4. Chicken Skin Mucosa – clusters of pale-yellow speckles in the surrounding mucosa (within ≈ 10 mm of the lesion margin). 5. Ulceration – a discrete mucosal defect exposing the subepithelial tissue on the lesion surface, usually covered by whitish fibrin/exudate (‘white plaque’) . 6. Spontaneous Bleeding – active oozing of blood from the lesion that is observed before any mechanical contact, irrigation or biopsy . Phase 2 – Establishing the Expert Consensus Gold Standard 20 images of LNPCPs were selected from a prospectively maintained tertiary referral center database to be broadly representative. The cases included 13 polyps without cancer, three with superficial SMI (< 1000µm), and four with deep SMI (≥ 1000µm). Four of these lesions had undergone prior manipulation (associated scarring) (Table 1 , Suppl Table 2 ). Each image displayed the polyp in white-light endoscopy at a distance of ≥ 10mm, without magnification or virtual chromoendoscopy (Olympus, Tokyo, Japan). Table 1 Demographic table . Demographics of selected colorectal polyps and demographic and experiential profile of survey participants. CI: confidence interval; IQR: interquartile range; pEMR: piecemeal endoscopic mucosal resection; SMI: submucosal invasive cancer. Colorectal Polyps, (n = 20) Size , median mm (95%CI) 29.5 (22.3–36.7) Morphology , n (%) non granular 12 (60) granular 7 (35) sessile serrated 1 (5) Paris Classification , n (%) Paris IIa 11 (55) Paris Is component 7 (35) Paris C component 2 (10) Location , n (%) rectum 4 (20) left colon 2 (10) right colon 14 (70) Histopathology , n (%) No Cancer 13 (65) Superficial SMI 3 (15) Deep SMI 4 (20) Participants , (n = 165) Profession , n (%) Consultant gastroenterologist 105 (63.6) Consultant surgeon 22 (13.3) Trainee gastroenterologist 32 (19.4) Trainee surgeon 3 (1.8) Medical student 3 (1.8) Years in practice , median [IQR] 6.5 [15.0] Lifetime colonoscopies ≥ 1000 procedures, n (%) 94 (57.0) Lifetime pEMRs ≥ 50 procedures, n (%) 21 (12.7) Continent of practice , n (%) Europe 119 (72.1) Africa 19 (11.5) Oceania 12 (7.3) Asia 15 (5.7) Americas 6 (3.6) Histopathological examination of all polyps was performed by an expert gastrointestinal pathologist (AH), blinded to the endoscopic assessment. The pathologist determined the presence and depth of cancer invasion, which served as the gold standard for Blink (cancer vs no cancer). Only cases with SMI were classified as “cancer”; cases with ‘intramucosal cancer’ were categorized as high-grade dysplasia and labelled “no cancer” for the purposes of this study. Four authors (DJT, JA, LDs, LDb), blinded to histopathology, independently assessed the images through an online platform. For each image the presence or absence of each Blink feature was recorded. Thereafter a structured consensus meeting was held to discuss discrepancies and establish a single gold standard for Blink feature assignment. Phase 3 – Validation of the Blink and Blink Features An original, purpose-built online survey ( Supplementary Survey) was developed using SurveyMonkey (California, USA) comprising four parts: 1. Demographics and Experience : Participants self-reported their role (consultant gastroenterologist, surgeon, trainee or medical student), years of experience, and procedural volumes (colonoscopies and piecemeal endoscopic mucosal resections [pEMRs]). General endoscopists were defined as those performing colonoscopy as part of routine clinical practice but without a recognised subspeciality focus in endoscopic imaging, optical diagnosis, or complex polyp resection (e.g., EMR/ESD). 2. Blink Assessment (Pre-Intervention) : Participants viewed 20 LNPCP images in a random order and recorded their Blink impression (presence of cancer: yes/no). Review was untimed. 3. The Educational Intervention : Participants were shown a 2-minute educational video[ 29 ], which introduced the six Blink features with definitions and brief annotated examples distinct from the survey images. 4. Blink Feature Identification (Post-Intervention) : Participants reviewed the same 20 images in a random order, recorded their Blink impression and the presence or absence of each of the six Blink features. Invitations to complete the survey were distributed via email to subscribers of the Gastrointestinal Quality and Safety (GIEQs) Foundation mailing list (Ghent, Belgium). An initial invitation was sent on 9 November 2022, followed by a reminder on 22 November 2022. The survey closed on 28 November 2022. Study Endpoints Primary Endpoint Change in miss rate for recognising histologically proven submucosal invasive cancer in LNPCPs using endoscopic imaging after exposure to six Blink features. Secondary Endpoints 1. Change in specificity for recognising cancer after Blink-feature training. 2. Predictive value of each individual Blink feature for discriminating cancer. 3. Association between Blink-feature count and histologically confirmed SMI. 4. Inter-observer agreement for presence/absence of each Blink feature. 5. Effect of endoscopist experience on diagnostic accuracy. Statistical Analysis All statistical analyses were performed using R statistical software (R Foundation, Indiana, USA) on anonymized data. Incomplete surveys were excluded from analysis. Demographic data are presented as percentages, medians, and interquartile ranges. Sensitivity and specificity were estimated using generalized linear mixed models (GLMMs) to account for multiple raters assessing the same polyps, with random effects for polyps and respondents[ 30 ]. Marginal estimates and pairwise contrasts were calculated using the emmeans package[ 31 ]. Stratified analyses assessed performance by experience level. A logistic mixed-effects model evaluated individual Blink features as predictors of cancer, with performance assessed via 5-fold cross-validation and ROC analysis. Full statistical methods are provided in Supplementary Methods. Results Study population 165 endoscopists from 52 countries completed the online survey, generating 3,300 ratings of 20 LNPCPs pre- and post the intervention. Most were consultant gastroenterologists (105; 63.6%) or surgeons (22; 13.3%); 35 (21.2%) were trainees and the remainder medical students (3; 1.8%). Median self-reported colonoscopy experience was 6.5 years (IQR 15.0). More than half (94; 57.0%) had performed > 1000 colonoscopies, and a smaller proportion (21; 12.7%) had completed at least 50 pEMRs. Europe predominated (119, 72.1%), but all continents were represented (Table 1 , Suppl Table 3 ). Impact of exposure to Six Blink features on diagnostic performance Performance before and after the two-minute educational video was compared (Table 2 ). At baseline, participants identified histologically proven cancer with 73.4% sensitivity (95% confidence interval [95%CI] 54.0–86.6) and 75.0% specificity (95%CI 61.5–84.9). After exposure to the concept of six Blink features, sensitivity rose to 94.3% (95%CI 87.2–97.6) – an absolute gain of 20.9% (95%CI 8.6–33.2; P < 0.001). The miss-rate fell four-fold (from 26.6% to 5.7%) with a fall in specificity to 57.8% (95%CI 41.8–72.3) and a rise in the false-alarm rate to 42.2% (95%CI 27.7–58.2). Table 2 Overall and feature-specific diagnostic performance of the Blink impression. The upper block presents accuracy metrics before and after the two-minute Blink-feature educational intervention, including sensitivity, specificity, miss rate (1 – sensitivity), and false alarm rate (i.e. false positive rate). The lower block displays per-Blink-feature diagnostic indices and their independent associations with cancer. CI: confidence interval. Baseline Post-intervention Absolute Δ P Global metrics Sensitivity, % (95% CI) 73.4 (54.0–86.6) 94.3 (87.2–97.6) + 20.9 (8.6–33.2) < 0.001 Miss-rate, % (95% CI) 26.6 (13.4–46.0) 5.7 (2.4–12.8) –20.9 (–33.2 to − 8.6) < 0.001 Specificity, % (95% CI) 75.0 (61.5–84.9) 57.8 (41.8–72.3) –17.2 (–22.9 to − 11.4) < 0.001 False-alarm rate, % (95% CI) 25.0 (15.1–38.5) 42.2 (27.7–58.2) + 17.2 (11.4–22.9) < 0.001 Individual cues Sensitivity, % (95% CI) Specificity, % (95% CI) Adjusted OR† (95% CI) P Extra redness 72.0 (60.9–81.0) 72.9 (64.9–79.7) 3.66 (3.09–4.33) < 0.001 Depression 64.4 (52.3–74.9) 71.1 (62.9–78.2) 3.06 (2.58–3.64) < 0.001 Fold deformation 44.5 (32.7–56.9) 71.7 (63.5–78.7) 1.18 (0.99–1.40) 0.060 Spontaneous bleeding 27.1 (18.3–38.0) 94.2 (91.6–96.0) 3.92 (3.11–4.96) < 0.001 Ulceration 21.3 (14.0–31.0) 92.5 (89.2–94.8) 1.11 (0.88–1.40) 0.359 Chicken-skin mucosa 17.2 (11.1–25.7) 89.1 (84.7–92.3) 1.03 (0.82–1.28) 0.822 Subgroup analyses revealed the same pattern irrespective of seniority, colonoscopy count, or EMR experience ( Suppl. Tables 4–6 ). Number of Blink-features and histological invasion depth The mean number of Blink features identified per polyp was plotted against final histology (Fig. 2 ). Benign lesions yielded mean 1.08 (95%CI 0.82–1.40) features from participant ratings versus 2.46 (95%CI 1.85–3.12) in those with invasive cancer. Using a threshold of ≥ 2 features yielded 78.2% (95%CI 90.8–59.9) sensitivity and 70.5% (95%CI 80.8–57.3) specificity for invasive cancer, while ≥ 3 features were more specific (92.4% [95%CI 96.3–85.5]) but only modestly sensitive (47.6% [95%CI 69.2–27.2]) ( Suppl. Table 7 ). Diagnostic yield of individual Blink features Extra redness was the most sensitive and specific (72.0%/72.9%) Blink feature for invasive cancer amongst the respondents, followed by depression (64.4%/71.1%) and fold deformation (44.5%/71.7%). Conversely, spontaneous bleeding and ulceration were rare but telling signs, each occurring in 92%. Chicken-skin mucosa was the least sensitive marker (17.2%) but had a high specificity of 89.1% (Table 2 ). Comparing post-intervention participant ratings (benign vs cancer) versus histopathology, true-negatives (TN) were the 1189 occasions a benign lesion was correctly dismissed, false-positives (FP) the 956 ratings where benign lesions were actually cancer, false-negatives (FN) the 137 ratings where cancers were missed and true-positives (TP) the 1018 ratings where cancers were correctly detected (Fig. 3 ). Three patterns emerged: (i) extra redness, depression, fold deformation dominated correct cancer calls: they appeared in 76%, 70%, and 48% of TP ratings but in ≤ 21% of FN ratings, (ii) specificity loss was driven by “over-calling” redness and fold deformation—both increased from 12–17% in TN to about 40–45% in FP decisions, (iii) rare yet vivid signs—spontaneous bleeding and ulceration—were almost absent in TN (≤ 2.4%) but were observed significantly more in TP (24–29%) (Table 3 ). Table 3 Prevalence of each Blink cue within diagnostic quadrants (3300 post-intervention ratings) . Each cell shows the percentage of ratings in which the Blink feature was marked present. Columns correspond to the four combinations of ground-truth histology and participant prediction. CI: confidence interval. True-negative (benign ↔ predicted benign) n = 1189 False-positive (benign ↔ predicted cancer) n = 956 False-negative (cancer ↔ predicted benign) n = 137 True-positive (cancer ↔ predicted cancer) n = 1018 Chicken-skin mucosa % (95% CI) 4.5 (3.2–6.3) 8.5 (4.7–15.0) 18.9 (14.4–24.5) 18.9 (14.4–24.5) Depression % (95% CI) 9.3 (7.0–12.1) 20.9 (13.7–30.5) 69.9 (62.7–76.2) 69.9 (62.7–76.2) Fold deformation % (95% CI) 16.5 (13.0–20.6) 20.2 (13.2–29.6) 47.9 (40.1–55.9) 47.9 (40.1–55.9) Extra redness % (95% CI) 12.3 (9.5–15.7) 39.9 (29.5–51.4) 75.7 (69.2–81.2) 75.7 (69.2–81.2) Spontaneous bleeding % (95% CI) 2.4 (1.6–3.7) 13.8 (8.4–21.9) 29.3 (23.1–36.4) 29.3 (23.1–36.4) Ulceration % (95% CI) 0.9 (0.5–1.7) 7.9 (4.3–14.1) 23.7 (18.3–30.1) 23.7 (18.3–30.1) Multivariable modelling The mixed-effects logistic model confirmed three features as independent predictors of general endoscopists detecting cancer within a given polyp (Table 2 ): spontaneous bleeding: OR 3.92 (95%CI 3.11–4.96), extra redness: OR 3.66 (3.09–4.33) and depression: OR 3.06 (2.58–3.64). Fold deformation trended towards significance (OR 1.18; P = 0.060), whereas chicken-skin mucosa and ulceration were not significant after adjustment. Model fit was excellent (AUC 0.79) ( Suppl. Table 8, Suppl. Figure 1 ), and the participant random effect collapsed to near-zero, indicating minimal unexplained between-rater heterogeneity once feature-level information was incorporated. Inter-observer agreement Raw post-intervention agreement for presence/absence of each feature ranged from 0.53 (depression) to 0.94 (spontaneous bleeding). Fleiss’ κ values were modest—0.14 for fold deformation, 0.51 for spontaneous bleeding. Intraclass correlation coefficients for the composite Blink score varied from 0.45 to 0.75 across features ( Suppl Table 9 ). Discussion This prospective, image-based interventional study demonstrates that the described 6 Blink features fundamentally address the performance gap of general endoscopists—those who perform colonoscopy routinely but lack expertise in endoscopic imaging—in detecting cancer within colorectal polyps. After a two-minute online introduction to the Blink-feature concept the cancer miss-rate fell four-fold. Importantly, the included images were taken using standard white-light endoscopy without magnification or virtual chromoendoscopy, making these improvements immediately accessible to the majority of endoscopists who lack advanced imaging expertise. The study's relevance lies not in enhancing expert performance—specialists in endoscopic imaging have already developed sophisticated unconscious Blink pattern recognition—but in democratizing cancer detection for the broader endoscopic community. Primary endpoint – a large, uniform gain in sensitivity Breaking down the first-impression of cancer into six Blink features delivered a 21% absolute rise in sensitivity amongst general endoscopists. The effect was consistent across consultants, trainees, and surgeons. The magnitude of this effect—reducing the miss rate from 27% to 6%—represents a substantial improvement in this controlled setting. If similar gains in the live environment could be made may prevent up to one in five cancers from being inappropriately managed endoscopically. The price of this heightened vigilance was a 17% fall in specificity (from 75% to 58%). This trade-off is familiar in screening medicine: more disease is captured at the expense of more benign lesions being labelled suspicious[ 32 ]. The clinical implications of this sensitivity-specificity balance require careful consideration within the colorectal polyp management pathway. Over-calling benign polyps should trigger established safety nets—multidisciplinary team review or referral for expert optical diagnosis—that effectively mitigate potential harm[ 33 ]. By contrast, the consequences of missing cancer are substantial and potentially irreversible: attempted pEMR with positive margins, perforation and tumor seeding, and, delay to definitive oncological therapy. Given these asymmetric consequences, the resultant sensitivity-specificity balance appears not only acceptable but potentially desirable for those endoscopists performing frontline assessments. Secondary endpoints To understand why participants succeeded or failed after training, we mapped every rating onto a 2×2 matrix of ground-truth histology (benign/cancer) versus participant verdict (benign/cancer). Three distinct patterns emerged from this analysis: 1. Blink features that secure correct cancer calls When cancers were correctly identified, extra redness, depression, and fold deformation were the commonest cues recognized by participants. These same features showed markedly different prevalence in false-negative decisions, suggesting that failure to recognize these signs—rather than their absence—drove most misclassifications. This pattern implies that further refinement of teaching materials could focus on threshold recognition: at what point does mild erythema become "extra redness," and when does surface irregularity constitute a true "depression". 2. Features that erode specificity False-positive escalation of benign polyps was largely driven by over-calling extra redness and fold deformation. The high false-positive rate for fold deformation may reflect a specific clinical scenario: scarring from previous incomplete resection can produce architectural distortion mimicking invasive growth. Teaching must therefore refine when these two signs truly imply invasion versus benign fibrosis, potentially by including number of folds converging within the polyp or combining with high specificity features. Chicken-skin mucosa added little independent information once the other cues were known in this dataset, suggesting potential redundancy in the current six-item framework 3. Rare but powerful signs Spontaneous bleeding and ulceration remained almost absent in true-negatives yet increased dramatically in true-positives, explaining their very high specificities and strong adjusted odds ratios in the multivariate model. These "high-value" signs function as near-pathognomonic markers of invasion when present, though their relative rarity in the dataset limited their contribution to overall sensitivity. Together these patterns both help us understand how cancer decisions are made by general endoscopists and hint at a targeted refinement for the next training iteration: raise awareness that there should be a higher diagnostic threshold/information for extra redness and fold deformation while reinforcing that spontaneous bleeding and ulceration, though uncommon, are high-yield indicators of malignancy. Seen differently these are precisely the situations where expert assessment is important; these features can identify areas of polyps which can be carefully interrogated with magnification and chromoendoscopy for pit/vascular pattern analysis to reach a precise diagnosis. However, the lesion must first be detected and escalated to the expert and so both strategies and structured communication between colleagues are therefore essential. The multivariate model's random-participant variance collapsed towards zero, indicating that once feature use was accounted for, little unexplained heterogeneity remained between endoscopists. Such a pattern strongly supports this structured training approach over experiential learning alone. Inter-observer agreement Despite increased sensitivity using the framework, Fleiss' κ for most features stayed in the slight-to-fair range, with only spontaneous bleeding achieving moderate agreement. This persistent inter-observer variability is not unique to our study[ 34 ]. Importantly, however, despite modest feature-level agreement the overall framework still produced substantial improvements in overall diagnostic performance, suggesting that its value lies in its overall outcome (number of features) rather than perfect concordance on individual features. Experience still helps, but structure helps more Amongst participants, those with extensive colonoscopy experience posted the highest post-training sensitivity, yet trainees showed the greatest absolute gain, underscoring that explicit deconstruction of a problem can compress learning curves. This differential improvement pattern reveals an important distinction: experience in performing colonoscopy does not automatically confer expertise in endoscopic imaging. Many experienced gastroenterologists lack formal training in optical diagnosis, while conversely, some junior endoscopists may have received structured imaging education. The Blink framework bridges this specific skill gap by making a certain amount of tacit imaging knowledge explicit, allowing general endoscopists to appropriate pattern recognition strategies typically reserved for imaging specialists. That said, specificity dropped equally across all experience strata, reinforcing that if considering improving upon these results in the future the challenge lies not in experience per se but in calibrating appropriate thresholds for suspicious features and criteria for expert referral via multi-disciplinary discussion. Grounding the findings in learning theory According to dual-process models of cognition[ 9 ], experts use fast, pattern-recognition "System 1" heuristics, while novices rely on slower, analytic "System 2" reasoning. Our data suggest that explicitly labelling key visual patterns (introducing Blink features framework) allows general endoscopists to appropriate elements of System 1 thinking almost immediately—a concept validated in parallel fields of medicine [ 10 , 11 ]. First, it reduces intrinsic cognitive load by chunking complex visual information into six discrete, memorable features. Second, it minimizes extraneous load by providing a consistent evaluation sequence. Third, it converts vague intuitions ("this looks worrying") into explicit, teachable schemas that can be rehearsed and automated through deliberate practice. This scaffolding is particularly valuable in endoscopy, where real-time decision-making under the cognitive pressure of a complex motor skill often precludes extensive deliberation. Strengths and limitations Strengths of this study include the large globally diverse cohort, a consensual expert gold standard with forced consensus methodology, pre-post design and mixed-effects statistics that appropriately account for clustering within participants. The image quality—obtained from standard-definition endoscopes with good preparation—likely represents real-life conditions for visualization. Several limitations warrant mention. First, our 20-image set was enriched for cancer (~ 35%), potentially boosting sensitivity relative to everyday practice; repeating the identical images after the learning-intervention also risks a recall effect (somewhat mitigated by randomization). Second, the small sample restricts statistical power and makes κ estimates imprecise. Third, still photographs omit dynamic cues (bleeding on contact, fold deformation during peristalsis) and the cognitive load of live colonoscopy, so translation to real-time performance remains unproven. Fourth, the study measures diagnostic accuracy only; we did not track downstream outcomes to show fewer interval cancers or avoided surgeries, nor did we quantify the additional referrals generated by the specificity drop. Fifth, voluntary, education-minded participants may overstate uptake compared with the broader endoscopy workforce. Finally, our binary benign-vs-cancer framework does not accommodate the clinical grey zone of indeterminate lesions. Future directions: “Blink first, then look closer” We propose a two-tier implementation pathway integrating rapid screening with detailed analysis. When an LNPCP is identified, the operator pauses the endoscopic image and performs a rapid assessment of the six Blink features. If two or more are present, this triggers mandatory escalation. In interventional endoscopy settings, this means proceeding to (virtual) chromoendoscopy and magnification to evaluate established criteria (e.g. vascular/pit pattern analysis). In screening examinations, high-quality images and video should be captured for multidisciplinary team review. This approach respects efficiency demands while providing a safety net against the documented false-positive tendency. For those seeking to optimize their specificity using this framework, threshold refinement through borderline case libraries—particularly for redness and fold deformation—may further enhance performance. However, the core value proposition remains sensitivity improvement for general endoscopists. The six Blink features also constitute human-interpretable labels that could seed explainable AI algorithms, enabling real-time decision support that transparently indicates which features triggered concern. Conclusion A concise six-item Blink framework reduced the miss rate for invasive cancer in LNPCP still-images amongst general endoscopists 4-fold, while revealing precisely which visual cues drive both successful detection and false alarms. Embedding this structured approach into training curricula, coupled with selective use of confirmatory imaging for screen-positive lesions could offer an evidence-based pathway to reduce missed cancers while managing the increased false-positive rate through existing clinical safeguards. Further validation in live endoscopy settings will determine whether these gains in still-image interpretation translate to meaningful improvements in real-world patient outcomes. Declarations Potential competing interests and funding statement DJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical. The remaining authors have no competing interests to disclose. The University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript. Ethics approval and consent to participate The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of Ghent University Hospital (ONZ-2022-0488). The study was registered on ClinicalTrials.gov (NCT05699954) on 20 April 2023. Informed consent was obtained from participating patients and endoscopists prior to the study. Consent for publication Not applicable. No individual participant data or identifiable images are included in this manuscript. Competing Interests DJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical.The remaining authors have no competing interests to disclose.The University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript. Funding The University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript. Author Contribution Lynn Debels Analysed data, wrote the manuscript and revised according to author comments Tamas Tornai Analysed data, assisted with writing and revision of manuscriptJohn Anderson Assisted with writing and revision of manuscript Maria Eva Argenziano Assisted with writing and revision of manuscript Anne Hoorens Histopathological analysisVikash Lala Assisted with writing and revision of manuscript Pieter Jan Poortmans Assisted with writing and revision of manuscript Roland Valori Assisted with writing and revision of manuscript Lobke Desomer Assisted with writing and revision of manuscript David J Tate Designed study, designed algorithm, performed procedures, collected data, analyzed data, wrote manuscript, revised manuscript after co-authors remarks Acknowledgement Not applicable Data Availability The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request. References Zauber AG, Winawer SJ, O'Brien MJ, et al. Colonoscopic polypectomy and long-term prevention of colorectal-cancer deaths. N Engl J Med. 2012;366:687–96. Moss A, Bourke MJ, Williams SJ, et al. Endoscopic mucosal resection outcomes and prediction of submucosal cancer from advanced colonic mucosal neoplasia. Gastroenterology. 2011;140:1909–18. 10.1053/j.gastro.2011.02.062 . Jayanna M, Burgess NG, Singh R, et al. Cost Analysis of Endoscopic Mucosal Resection vs Surgery for Large Laterally Spreading Colorectal Lesions. Clin Gastroenterol hepatology: official Clin Pract J Am Gastroenterological Association. 2016;14:271–2. Keswani RN, Law R, Ciolino JD, et al. Adverse events after surgery for nonmalignant colon polyps are common and associated with increased length of stay and costs. Gastrointest Endosc. 2016;84:296–e303291. Saito Y, Sakamoto T, Dekker E, et al. First report from the International Evaluation of Endoscopic classification Japan NBI Expert Team: International multicenter web trial. Dig Endosc. 2023. 10.1111/den.14682 . Dekker E, Houwen B, Puig I, et al. Curriculum for optical diagnosis training in Europe: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy. 2020;52:899–923. 10.1055/a-1231-5123 . Meulen LWT, van de Wetering AJP, Debeuf MPH, et al. Optical diagnosis of T1 CRCs and treatment consequences in the Dutch CRC screening programme. Gut. 2020;69:2049–51. 10.1136/gutjnl-2019-320403 . Meulen LWT, Haasnoot KJC, Vlug MS, et al. Effect of optical diagnosis training on recognition and treatment of submucosal invasive colorectal cancer in community hospitals: a prospective multicenter intervention study. Endoscopy. 2024;56:770–9. 10.1055/a-2313-4996 . Kahneman Da. Thinking, fast and slow. 1st ed. New York: Farrar, Straus and Giroux, [2011] ©2011; 2011. Raat EM, Farr I, Wolfe JM, et al. Comparable prediction of breast cancer risk from a glimpse or a first impression of a mammogram. Cogn Res Princ Implic. 2021;6:72. 10.1186/s41235-021-00339-5 . Evans KK, Georgian-Smith D, Tambouret R, et al. The gist of the abnormal: above-chance medical decision making in the blink of an eye. Psychon Bull Rev. 2013;20:1170–5. 10.3758/s13423-013-0459-3 . Puig I, Mármol C, Bustamante M. Endoscopic imaging techniques for detecting early colorectal cancer. Curr Opin Gastroenterol. 2019;35:432–9. 10.1097/mog.0000000000000570 . Bugajski M, Kaminski MF, Orlowska J, et al. Suspicious macroscopic features of small malignant colorectal polyps. Scand J Gastroenterol. 2015;50:1261–7. 10.3109/00365521.2015.1024280 . Guan J, Zhao R, Zhang X, et al. Chicken skin mucosa surrounding adult colorectal adenomas is a risk factor for carcinogenesis. Am J Clin Oncol. 2012;35:527–32. 10.1097/COC.0b013e31821dedf7 . Horie H, Togashi K, Kawamura YJ, et al. Colonoscopic stigmata of 1 mm or deeper submucosal invasion in colorectal cancer. Dis Colon Rectum. 2008;51:1529–34. 10.1007/s10350-008-9263-y . Ikehara H, Saito Y, Matsuda T, et al. Diagnosis of depth of invasion for early colorectal cancer using magnifying colonoscopy. J Gastroenterol Hepatol. 2010;25:905–12. 10.1111/j.1440-1746.2010.06275.x . Jang HW, Park SJ, Cheon JH, et al. Does magnifying narrow-band imaging or magnifying chromoendoscopy help experienced endoscopists assess invasion depth of large sessile and flat polyps? Dig Dis Sci. 2014;59:1520–8. 10.1007/s10620-014-3090-x . Kudo S, Hirota S, Nakajima T, et al. Colorectal tumours and pit pattern. J Clin Pathol. 1994;47:880–5. 10.1136/jcp.47.10.880 . Kudo S, Kashida H, Tamura T, et al. Colonoscopic diagnosis and management of nonpolypoid early colorectal cancer. World J Surg. 2000;24:1081–90. 10.1007/s002680010154 . Lee YM, Song KH, Koo HS, et al. Colonic Chicken Skin Mucosa Surrounding Colon Polyps Is an Endoscopic Predictive Marker for Colonic Neoplastic Polyps. Gut Liver. 2022;16:754–63. 10.5009/gnl210271 . Matsuda T, Saito Y, Hotta K, et al. Prevalence and clinicopathological features of nonpolypoid colorectal neoplasms: should we pay more attention to identifying flat and depressed lesions? Dig Endosc. 2010;22(Suppl 1):S57–62. 10.1111/j.1443-1661.2010.00967.x . Misawa M, Kudo SE, Wada Y, et al. Magnifying narrow-band imaging of surface patterns for diagnosing colorectal cancer. Oncol Rep. 2013;30:350–6. 10.3892/or.2013.2471 . Nowicki MJ, Bishop PR, Subramony C, et al. Colonic chicken-skin mucosa in children with polyps is not a preneoplastic lesion. J Pediatr Gastroenterol Nutr. 2005;41:600–6. 10.1097/01.mpg.0000179658.09210.b6 . Puig I, López-Cerón M, Arnau A, et al. Accuracy of the Narrow-Band Imaging International Colorectal Endoscopic Classification System in Identification of Deep Invasion in Colorectal Polyps. Gastroenterology. 2019;156:75–87. 10.1053/j.gastro.2018.10.004 . Saito Y, Fujii T, Kondo H, et al. Endoscopic treatment for laterally spreading tumors in the colon. Endoscopy. 2001;33:682–6. 10.1055/s-2001-16213 . Saitoh Y, Obara T, Watari J, et al. Invasion depth diagnosis of depressed type early colorectal cancers by combined use of videoendoscopy and chromoendoscopy. Gastrointest Endosc. 1998;48:362–70. 10.1016/s0016-5107(98)70004-5 . Shatz BA, Weinstock LB, Thyssen EP, et al. Colonic chicken skin mucosa: an endoscopic and histological abnormality adjacent to colonic neoplasms. Am J Gastroenterol. 1998;93:623–7. 10.1111/j.1572-0241.1998.177_b.x . Uno Y, Munakata A. Endoscopic and histologic correlates of colorectal polyp bleeding. Gastrointest Endosc. 1995;41:460–7. 10.1016/s0016-5107(05)80004-5 . Tate DJ. Introduction to 6 Blink Features - Learning Tool. In: Online: 2022: https://vimeo.com/768168090 Bates D, Mächler M, Bolker B, et al. Fitting Linear Mixed-Effects Models Using lme4. J Stat Softw. 2015;67:1–48. 10.18637/jss.v067.i01 . emmeans RL. Estimated Marginal Means, aka Least-Squares Means. R package version 1.11.2-00002,. 2025. Shaughnessy AF. Clinical Epidemiology: A Basic Science for Clinical Medicine. BMJ. 2007;335:777. Di Fabio F, Jitsumura M, Longstaff L, et al. Management of Significant Polyp and Early Colorectal Cancer Is Optimized by Implementation of a Dedicated Multidisciplinary Team Meeting: Lessons Learned From the United Kingdom National Program. Dis Colon Rectum. 2022;65:654–62. 10.1097/dcr.0000000000002199 . van Doorn SC, Hazewinkel Y, East JE, et al. Polyp morphology: an interobserver evaluation for the Paris classification among international experts. Am J Gastroenterol. 2015;110:180–7. 10.1038/ajg.2014.326 . Additional Declarations Competing interest reported. DJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical. The remaining authors have no competing interests to disclose. The University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript. Supplementary Files BlinkSuppl.docx Supplementarysurvey.pdf floatimage1.jpeg Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8553404","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":591777571,"identity":"e5c0a0a4-3b2e-4d3d-97ae-ad890b2c3e34","order_by":0,"name":"Lynn K. Debels","email":"","orcid":"","institution":"University Hospital Brussels (UZ Brussel)","correspondingAuthor":false,"prefix":"","firstName":"Lynn","middleName":"K.","lastName":"Debels","suffix":""},{"id":591777572,"identity":"5a0859db-042b-474e-ae25-4516dde4fb49","order_by":1,"name":"Tamas Tornai","email":"","orcid":"","institution":"University Hospital Ghent (UZ Gent)","correspondingAuthor":false,"prefix":"","firstName":"Tamas","middleName":"","lastName":"Tornai","suffix":""},{"id":591777573,"identity":"60dbc72f-db54-471b-9b9c-ea4afa79071e","order_by":2,"name":"John Anderson","email":"","orcid":"","institution":"Gloucestershire Hospitals NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Anderson","suffix":""},{"id":591777574,"identity":"91f23aa3-70dd-4261-b6bb-41cf0170acb3","order_by":3,"name":"Maria Eva Argenziano","email":"","orcid":"","institution":"Ghent University","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Eva","lastName":"Argenziano","suffix":""},{"id":591777575,"identity":"f66150a8-9f60-4e21-90ea-d64ec062d5f2","order_by":4,"name":"Lobke Desomer","email":"","orcid":"","institution":"Ghent University","correspondingAuthor":false,"prefix":"","firstName":"Lobke","middleName":"","lastName":"Desomer","suffix":""},{"id":591777576,"identity":"19793a7a-7add-4e31-b58e-c75c8e967c6d","order_by":5,"name":"Anne Hoorens","email":"","orcid":"","institution":"University Hospital Ghent (UZ Gent)","correspondingAuthor":false,"prefix":"","firstName":"Anne","middleName":"","lastName":"Hoorens","suffix":""},{"id":591777577,"identity":"8a62a8bc-168f-41a7-9104-1a97fbbb3e5b","order_by":6,"name":"Vikash Lala","email":"","orcid":"","institution":"Charlotte Maxeke Johannesburg Academic Hospital","correspondingAuthor":false,"prefix":"","firstName":"Vikash","middleName":"","lastName":"Lala","suffix":""},{"id":591777578,"identity":"b8a066c4-0647-4cf4-9023-b0c36eae03dd","order_by":7,"name":"Pieter Jan Poortmans","email":"","orcid":"","institution":"University Hospital Brussels (UZ Brussel)","correspondingAuthor":false,"prefix":"","firstName":"Pieter","middleName":"Jan","lastName":"Poortmans","suffix":""},{"id":591777579,"identity":"800ba806-6f28-49f7-916c-3fb740660d0c","order_by":8,"name":"Roland Valori","email":"","orcid":"","institution":"Gloucestershire Hospitals NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Roland","middleName":"","lastName":"Valori","suffix":""},{"id":591777580,"identity":"aaef2fcc-3370-4df1-9802-0aa6f6464aae","order_by":9,"name":"David James Tate","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA70lEQVRIiWNgGAWjYDCCA0DEA2YxH4AK8RClxQCoji0BqoEILQwQLTwGxGnhO37G8MAbhj/y9uw93yR//LLJs2fgPfgAnxbJMzkGB+cwGBj28JzdJiHZl1bMw8CXbIBPi8GBtITDQIcx9kjkbpMw7Dmc2MPAYyaBV8v5Z2At9j3yb55JJPb8B2kx/4FXy43kAyAtiT0SPGwSB34cANuCTweD5I3HBw7OMTBO7jmTZmzZ2JCc2HOYLxmvw/jOJzZ/eFMhZ9vefvjhzR9/7BLb23sPfsBrDcR5UJqxDUgwE1aPDP6QpnwUjIJRMApGBgAAyNdMNkG8nFgAAAAASUVORK5CYII=","orcid":"","institution":"Ghent University","correspondingAuthor":true,"prefix":"","firstName":"David","middleName":"James","lastName":"Tate","suffix":""}],"badges":[],"createdAt":"2026-01-08 15:54:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8553404/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8553404/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102853961,"identity":"8ebe1f39-c757-476b-9fa6-36750d48a0fa","added_by":"auto","created_at":"2026-02-17 14:46:33","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":197408,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRepresentative Images of each of the Six Blink features. \u003c/strong\u003eWhite-light endoscopic still images, each captured at ≥10 mm distance, demonstrate the six gross morphological “Blink” features: spontaneous bleeding; extra redness; depression; fold deformation; ulceration; chicken-skin mucosa. All features are marked by arrowheads. All images were acquired with standard-definition white-light endoscopy and without magnification or virtual chromoendoscopy. Odds ratios (OR) for cancer detection amongst general endoscopists with 95% confidence intervals (CI) are shown per Blink feature. Statistically significant results (p ≤ 0.05) are indicated in bold.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/bc32a02244f2b84283825e88.jpg"},{"id":102853960,"identity":"a7343e86-b73e-409e-a06d-4a6b12cbb92e","added_by":"auto","created_at":"2026-02-17 14:46:33","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":214418,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCorrelation between Blink-feature burden and histology.\u003c/strong\u003eDot-plot of rater level aggregated individual ratings\u003cstrong\u003e \u003c/strong\u003e(grey circles) showing the number of Blink features identified per rater (ordinate, 0–6) stratified by final histology (Benign vs Cancer). Large red circles represent the group mean and red lines the 95% confidence intervals. The shift from left to the right in the figure illustrates the increased Blink-feature burden in malignant lesions (linear-trend P \u0026lt; 0.001).\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/856c080b2c9bf913ded97864.jpg"},{"id":102853965,"identity":"9159290c-9d10-43e8-a25a-b10ac588d087","added_by":"auto","created_at":"2026-02-17 14:46:34","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":211587,"visible":true,"origin":"","legend":"\u003cp\u003ePercentage of specific Blink features present in general endoscopist polyp ratings, stratified by true histological and predicted histologic status. Number of observations (predicted benign, true benign: 1189; predicted cancer, true benign: 956; predicted benign, true cancer: 137; predicted cancer, true cancer: 1018). Circles represent estimated marginal means; error bars indicate 95% confidence intervals.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/407d21db06b043aab00ea1d6.jpg"},{"id":104882299,"identity":"a76754a5-d567-42c1-a20f-a7ddc727975f","added_by":"auto","created_at":"2026-03-18 09:29:12","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2003138,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/9273ac27-bb1e-43d2-b463-c06ee5559a5e.pdf"},{"id":102853963,"identity":"b84e3d31-087c-44d4-8a87-ebf8538facbe","added_by":"auto","created_at":"2026-02-17 14:46:33","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":332319,"visible":true,"origin":"","legend":"","description":"","filename":"BlinkSuppl.docx","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/8ddecf85dd1b3d3c2960f647.docx"},{"id":102853966,"identity":"5fd97ca7-791b-41b7-aa6a-84a17c448343","added_by":"auto","created_at":"2026-02-17 14:46:35","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":51307130,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarysurvey.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/9582e9930c2500f7a42fb12c.pdf"},{"id":102853962,"identity":"288a9550-4ecd-4e06-b221-641207a36270","added_by":"auto","created_at":"2026-02-17 14:46:33","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":130700,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8553404/v1/425916b3f2b69837ab7d96be.jpeg"}],"financialInterests":"Competing interest reported. DJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical.\nThe remaining authors have no competing interests to disclose.\nThe University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript.","formattedTitle":"Teaching Six ‘Blink’ Features Reduces General Endoscopists’ Cancer Miss-Rate in Image-Based Assessment of Large Colorectal Polyps","fulltext":[{"header":"Introduction","content":"\u003cp\u003eColorectal cancer is one of the most preventable malignancies through the early detection and removal of precancerous lesions via polypectomy[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The technique required to endoscopically resect a given polyp is related to the presence of cancer within that lesion and the estimated depth of any submucosal invasion (SMI)[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Incorrect decision making can lead to morbidity and potential mortality for patients and costs for healthcare systems[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eReal-time polyp assessment for cancer risk using high-resolution endoscopy with magnification and virtual chromoendoscopy is effective amongst experts[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], but remains complex for general endoscopists without specific training[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Since these practitioners will perform the majority of procedures where large (\u0026ge;\u0026thinsp;10mm) non-pedunculated colorectal polyps (LNPCPs) are first detected there is a pressing need for simplified and robust approaches to allow estimation of malignant potential and effective triage of treatment strategy.\u003c/p\u003e \u003cp\u003eIt is well recognized that human experts, regardless of the domain of expertise, are exceptionally adept at synthesizing large volumes of visual, contextual, and historical information to arrive at a first (or \u0026ldquo;Blink\u0026rdquo;) impression[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. In medicine there is precedent for the Blink concept. One study[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] demonstrated that expert radiologists who received only brief exposure to mammograms could detect breast cancer as accurately as when they were allowed unlimited exposure to the same mammograms. In another study[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], both radiologists and cytologists were able to reliably stratify medical images as normal/abnormal during brief exposures (250\u0026ndash;2000 milliseconds) without being able to accurately localise the actual pathology during that exposure. Importantly both studies demonstrated that non-experts could not reliably detect the outcome based on brief exposures.\u003c/p\u003e \u003cp\u003eApplied to LNPCPs, the success of expert Blink impression likely hinges on the ability to quickly recognize specific visual features that are strongly associated with underlying malignancy when first encountering the polyp. It follows that these cues would be morphologic features visible from afar and without virtual chromoendoscopy, since there would be no time for detailed analysis of pit/vascular pattern within the required timeframe. So-called gross morphologic features, shown to significantly increase the risk of a polyp containing deep submucosally invasive cancer when identified by experts[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], would be excellent candidates.\u003c/p\u003e \u003cp\u003eThe authors proposed a condensed set of gross morphological characteristics which could potentially be recognized at-a-glance by general endoscopists, mimicking the intuitive pattern matching used by experts. It was hypothesized that such an approach could shortcut the need for significant experience to identify cancer in LNPCPs. The six selected features, hereafter referred to as \u0026ldquo;Blink features\u0026rdquo;, were spontaneous bleeding, depression, fold deformation, extra redness, ulceration and chicken-skin mucosa [\u003cspan additionalcitationids=\"CR14 CR15 CR16 CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, \u003cb\u003eSuppl. Materials \u0026ndash; Narrative Review\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study aimed to assess the impact of introducing a structured assessment of the six Blink features using a 2-minute learning intervention on the diagnostic performance of cancer prediction from still images of LNPCPs amongst general endoscopists.\u003c/p\u003e"},{"header":"Methods","content":"\n\u003ch3\u003eStudy Design\u003c/h3\u003e\n\u003cp\u003eThis was a prospective interventional study designed to evaluate the accuracy of colorectal cancer detection using Blink impression before and after the introduction of six Blink features using a 2-minute online learning intervention. The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of Ghent University Hospital (ONZ-2022-0488). The study was registered on ClinicalTrials.gov (NCT05699954) on 20 April 2023. Informed consent was obtained from participating patients and endoscopists prior to the study.\u003c/p\u003e \u003cp\u003ePhase 1 \u0026ndash; Selection of Blink Features\u003c/p\u003e \u003cp\u003eA comprehensive literature review was conducted in Medline, Embase, PubMed, and the Cochrane Library to identify gross morphological features of colorectal polyps associated with deep SMI (\u003cb\u003eSupplementary Materials\u003c/b\u003e, \u003cb\u003eSuppl. Table\u0026nbsp;1\u003c/b\u003e). From the identified literature, six features were selected by author consensus (DJT, JA, LDs, LDb) based on three criteria: (1) strength of evidence linking the feature to deep SMI, (2) clinical relevance and frequency of observation, and (3) ease of identification by endoscopists of varying experience using white-light endoscopy without magnification or image enhancement. The selected features, collectively referred to as the six Blink features were:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e1. \u003cb\u003eFold Deformation\u003c/b\u003e \u0026ndash; \u003cem\u003econvergence of \u0026ge;\u0026thinsp;3 adjacent haustral folds towards, or interruption of an existing fold by, the lesion.\u003c/em\u003e\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e2. \u003cb\u003eExtra Redness\u003c/b\u003e \u0026ndash; \u003cem\u003eone or more foci of mucosal erythema on the lesion surface that are a deeper hue than the immediately-adjacent polyp tissue\u003c/em\u003e.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e3. \u003cb\u003eDepression\u003c/b\u003e \u0026ndash; \u003cem\u003ea clearly demarcated concavity or excavated area on the luminal surface of the polyp\u003c/em\u003e .\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e4. \u003cb\u003eChicken Skin Mucosa\u003c/b\u003e \u0026ndash; \u003cem\u003eclusters of pale-yellow speckles in the surrounding mucosa (within \u0026asymp;\u0026thinsp;10 mm of the lesion margin).\u003c/em\u003e\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e5. \u003cb\u003eUlceration\u003c/b\u003e \u0026ndash; \u003cem\u003ea discrete mucosal defect exposing the subepithelial tissue on the lesion surface, usually covered by whitish fibrin/exudate (\u0026lsquo;white plaque\u0026rsquo;)\u003c/em\u003e.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e6. \u003cb\u003eSpontaneous Bleeding\u003c/b\u003e \u0026ndash; \u003cem\u003eactive oozing of blood from the lesion that is observed before any mechanical contact, irrigation or biopsy\u003c/em\u003e.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003ePhase 2 \u0026ndash; Establishing the Expert Consensus Gold Standard\u003c/p\u003e \u003cp\u003e20 images of LNPCPs were selected from a prospectively maintained tertiary referral center database to be broadly representative. The cases included 13 polyps without cancer, three with superficial SMI (\u0026lt;\u0026thinsp;1000\u0026micro;m), and four with deep SMI (\u0026ge;\u0026thinsp;1000\u0026micro;m). Four of these lesions had undergone prior manipulation (associated scarring) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, \u003cb\u003eSuppl Table\u0026nbsp;2\u003c/b\u003e). Each image displayed the polyp in white-light endoscopy at a distance of \u0026ge;\u0026thinsp;10mm, without magnification or virtual chromoendoscopy (Olympus, Tokyo, Japan).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eDemographic table\u003c/b\u003e. Demographics of selected colorectal polyps and demographic and experiential profile of survey participants. CI: confidence interval; IQR: interquartile range; pEMR: piecemeal endoscopic mucosal resection; SMI: submucosal invasive cancer.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eColorectal Polyps, (n\u0026thinsp;=\u0026thinsp;20)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSize\u003c/b\u003e, median mm (95%CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29.5 (22.3\u0026ndash;36.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMorphology\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enon granular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12 (60)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003egranular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (35)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esessile serrated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eParis Classification\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParis IIa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11 (55)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParis Is component\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (35)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParis C component\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (10)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLocation\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003erectum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (20)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eleft colon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (10)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eright colon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e14 (70)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistopathology\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo Cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13 (65)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSuperficial SMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (15)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeep SMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (20)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eParticipants\u003c/b\u003e, (n\u0026thinsp;=\u0026thinsp;165)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eProfession\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConsultant gastroenterologist\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e105 (63.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConsultant surgeon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22 (13.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrainee gastroenterologist\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e32 (19.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrainee surgeon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (1.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedical student\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (1.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYears in practice\u003c/b\u003e, median [IQR]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6.5 [15.0]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLifetime colonoscopies\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u0026ge;\u0026thinsp;1000 procedures, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e94 (57.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLifetime pEMRs\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u0026ge;\u0026thinsp;50 procedures, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21 (12.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eContinent of practice\u003c/b\u003e, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEurope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e119 (72.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAfrica\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19 (11.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOceania\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12 (7.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAsia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15 (5.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAmericas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (3.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eHistopathological examination of all polyps was performed by an expert gastrointestinal pathologist (AH), blinded to the endoscopic assessment. The pathologist determined the presence and depth of cancer invasion, which served as the gold standard for Blink (cancer vs no cancer). Only cases with SMI were classified as \u0026ldquo;cancer\u0026rdquo;; cases with \u0026lsquo;intramucosal cancer\u0026rsquo; were categorized as high-grade dysplasia and labelled \u0026ldquo;no cancer\u0026rdquo; for the purposes of this study.\u003c/p\u003e \u003cp\u003eFour authors (DJT, JA, LDs, LDb), blinded to histopathology, independently assessed the images through an online platform. For each image the presence or absence of each Blink feature was recorded. Thereafter a structured consensus meeting was held to discuss discrepancies and establish a single gold standard for Blink feature assignment.\u003c/p\u003e \u003cp\u003ePhase 3 \u0026ndash; Validation of the Blink and Blink Features\u003c/p\u003e \u003cp\u003eAn original, purpose-built online survey (\u003cb\u003eSupplementary Survey)\u003c/b\u003e was developed using SurveyMonkey (California, USA) comprising four parts:\u003c/p\u003e \u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003e1. Demographics and Experience\u003c/b\u003e: Participants self-reported their role (consultant gastroenterologist, surgeon, trainee or medical student), years of experience, and procedural volumes (colonoscopies and piecemeal endoscopic mucosal resections [pEMRs]). General endoscopists were defined as those performing colonoscopy as part of routine clinical practice but without a recognised subspeciality focus in endoscopic imaging, optical diagnosis, or complex polyp resection (e.g., EMR/ESD).\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003e2. Blink Assessment (Pre-Intervention)\u003c/b\u003e: Participants viewed 20 LNPCP images in a random order and recorded their Blink impression (presence of cancer: yes/no). Review was untimed.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003e3. The Educational Intervention\u003c/b\u003e: Participants were shown a 2-minute educational video[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], which introduced the six Blink features with definitions and brief annotated examples distinct from the survey images.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e4. \u003cb\u003eBlink Feature Identification (Post-Intervention)\u003c/b\u003e: Participants reviewed the same 20 images in a random order, recorded their Blink impression and the presence or absence of each of the six Blink features.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e \u003cp\u003eInvitations to complete the survey were distributed via email to subscribers of the Gastrointestinal Quality and Safety (GIEQs) Foundation mailing list (Ghent, Belgium). An initial invitation was sent on 9 November 2022, followed by a reminder on 22 November 2022. The survey closed on 28 November 2022.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Endpoints\u003c/h2\u003e \u003cp\u003ePrimary Endpoint\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eChange in miss rate for recognising histologically proven submucosal invasive cancer in LNPCPs using endoscopic imaging after exposure to six Blink features.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eSecondary Endpoints\u003c/p\u003e \u003cp\u003e1. Change in specificity for recognising cancer after Blink-feature training.\u003c/p\u003e \u003cp\u003e2. Predictive value of each individual Blink feature for discriminating cancer.\u003c/p\u003e \u003cp\u003e3. Association between Blink-feature count and histologically confirmed SMI.\u003c/p\u003e \u003cp\u003e4. Inter-observer agreement for presence/absence of each Blink feature.\u003c/p\u003e \u003cp\u003e5. Effect of endoscopist experience on diagnostic accuracy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eAll statistical analyses were performed using R statistical software (R Foundation, Indiana, USA) on anonymized data. Incomplete surveys were excluded from analysis.\u003c/p\u003e \u003cp\u003eDemographic data are presented as percentages, medians, and interquartile ranges. Sensitivity and specificity were estimated using generalized linear mixed models (GLMMs) to account for multiple raters assessing the same polyps, with random effects for polyps and respondents[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Marginal estimates and pairwise contrasts were calculated using the emmeans package[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Stratified analyses assessed performance by experience level. A logistic mixed-effects model evaluated individual Blink features as predictors of cancer, with performance assessed via 5-fold cross-validation and ROC analysis. Full statistical methods are provided in Supplementary Methods.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eStudy population\u003c/p\u003e \u003cp\u003e165 endoscopists from 52 countries completed the online survey, generating 3,300 ratings of 20 LNPCPs pre- and post the intervention. Most were consultant gastroenterologists (105; 63.6%) or surgeons (22; 13.3%); 35 (21.2%) were trainees and the remainder medical students (3; 1.8%). Median self-reported colonoscopy experience was 6.5 years (IQR 15.0). More than half (94; 57.0%) had performed\u0026thinsp;\u0026gt;\u0026thinsp;1000 colonoscopies, and a smaller proportion (21; 12.7%) had completed at least 50 pEMRs. Europe predominated (119, 72.1%), but all continents were represented (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, \u003cb\u003eSuppl Table\u0026nbsp;3\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eImpact of exposure to Six Blink features on diagnostic performance\u003c/p\u003e \u003cp\u003ePerformance before and after the two-minute educational video was compared (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). At baseline, participants identified histologically proven cancer with 73.4% sensitivity (95% confidence interval [95%CI] 54.0\u0026ndash;86.6) and 75.0% specificity (95%CI 61.5\u0026ndash;84.9). After exposure to the concept of six Blink features, sensitivity rose to 94.3% (95%CI 87.2\u0026ndash;97.6) \u0026ndash; an absolute gain of 20.9% (95%CI 8.6\u0026ndash;33.2; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The miss-rate fell four-fold (from 26.6% to 5.7%) with a fall in specificity to 57.8% (95%CI 41.8\u0026ndash;72.3) and a rise in the false-alarm rate to 42.2% (95%CI 27.7\u0026ndash;58.2).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u0026ensp;Overall and feature-specific diagnostic performance of the Blink impression. The upper block presents accuracy metrics before and after the two-minute Blink-feature educational intervention, including sensitivity, specificity, miss rate (1 \u0026ndash; sensitivity), and false alarm rate (i.e. false positive rate). The lower block displays per-Blink-feature diagnostic indices and their independent associations with cancer. CI: confidence interval.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBaseline\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePost-intervention\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAbsolute Δ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGlobal metrics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensitivity, %\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e73.4\u003c/p\u003e \u003cp\u003e(54.0\u0026ndash;86.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e94.3\u003c/p\u003e \u003cp\u003e(87.2\u0026ndash;97.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;20.9\u003c/p\u003e \u003cp\u003e(8.6\u0026ndash;33.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiss-rate, %\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26.6\u003c/p\u003e \u003cp\u003e(13.4\u0026ndash;46.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.7\u003c/p\u003e \u003cp\u003e(2.4\u0026ndash;12.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026ndash;20.9\u003c/p\u003e \u003cp\u003e(\u0026ndash;33.2 to \u0026minus;\u0026thinsp;8.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecificity, %\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e75.0\u003c/p\u003e \u003cp\u003e(61.5\u0026ndash;84.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e57.8\u003c/p\u003e \u003cp\u003e(41.8\u0026ndash;72.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026ndash;17.2\u003c/p\u003e \u003cp\u003e(\u0026ndash;22.9 to \u0026minus;\u0026thinsp;11.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFalse-alarm rate, %\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.0\u003c/p\u003e \u003cp\u003e(15.1\u0026ndash;38.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e42.2\u003c/p\u003e \u003cp\u003e(27.7\u0026ndash;58.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;17.2\u003c/p\u003e \u003cp\u003e(11.4\u0026ndash;22.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIndividual cues\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eSensitivity, %\u003c/b\u003e\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003eSpecificity, %\u003c/b\u003e\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eAdjusted OR\u0026dagger;\u003c/b\u003e\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExtra redness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e72.0\u003c/p\u003e \u003cp\u003e(60.9\u0026ndash;81.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e72.9\u003c/p\u003e \u003cp\u003e(64.9\u0026ndash;79.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.66\u003c/p\u003e \u003cp\u003e(3.09\u0026ndash;4.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDepression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e64.4\u003c/p\u003e \u003cp\u003e(52.3\u0026ndash;74.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71.1\u003c/p\u003e \u003cp\u003e(62.9\u0026ndash;78.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.06\u003c/p\u003e \u003cp\u003e(2.58\u0026ndash;3.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFold deformation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44.5\u003c/p\u003e \u003cp\u003e(32.7\u0026ndash;56.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71.7\u003c/p\u003e \u003cp\u003e(63.5\u0026ndash;78.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.18\u003c/p\u003e \u003cp\u003e(0.99\u0026ndash;1.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.060\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpontaneous bleeding\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.1\u003c/p\u003e \u003cp\u003e(18.3\u0026ndash;38.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e94.2\u003c/p\u003e \u003cp\u003e(91.6\u0026ndash;96.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.92\u003c/p\u003e \u003cp\u003e(3.11\u0026ndash;4.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUlceration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.3\u003c/p\u003e \u003cp\u003e(14.0\u0026ndash;31.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.5\u003c/p\u003e \u003cp\u003e(89.2\u0026ndash;94.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.11\u003c/p\u003e \u003cp\u003e(0.88\u0026ndash;1.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.359\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChicken-skin mucosa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17.2\u003c/p\u003e \u003cp\u003e(11.1\u0026ndash;25.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e89.1\u003c/p\u003e \u003cp\u003e(84.7\u0026ndash;92.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.03\u003c/p\u003e \u003cp\u003e(0.82\u0026ndash;1.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.822\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eSubgroup analyses revealed the same pattern irrespective of seniority, colonoscopy count, or EMR experience (\u003cb\u003eSuppl. Tables\u0026nbsp;4\u0026ndash;6\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eNumber of Blink-features and histological invasion depth\u003c/p\u003e \u003cp\u003eThe mean number of Blink features identified per polyp was plotted against final histology (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Benign lesions yielded mean 1.08 (95%CI 0.82\u0026ndash;1.40) features from participant ratings versus 2.46 (95%CI 1.85\u0026ndash;3.12) in those with invasive cancer. Using a threshold of \u0026ge;\u0026thinsp;2 features yielded 78.2% (95%CI 90.8\u0026ndash;59.9) sensitivity and 70.5% (95%CI 80.8\u0026ndash;57.3) specificity for invasive cancer, while\u0026thinsp;\u0026ge;\u0026thinsp;3 features were more specific (92.4% [95%CI 96.3\u0026ndash;85.5]) but only modestly sensitive (47.6% [95%CI 69.2\u0026ndash;27.2]) (\u003cb\u003eSuppl. Table\u0026nbsp;7\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eDiagnostic yield of individual Blink features\u003c/p\u003e \u003cp\u003eExtra redness was the most sensitive and specific (72.0%/72.9%) Blink feature for invasive cancer amongst the respondents, followed by depression (64.4%/71.1%) and fold deformation (44.5%/71.7%). Conversely, spontaneous bleeding and ulceration were rare but telling signs, each occurring in \u0026lt;\u0026thinsp;30% of cancers yet with specificities\u0026thinsp;\u0026gt;\u0026thinsp;92%. Chicken-skin mucosa was the least sensitive marker (17.2%) but had a high specificity of 89.1% (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eComparing post-intervention participant ratings (benign vs cancer) versus histopathology, true-negatives (TN) were the 1189 occasions a benign lesion was correctly dismissed, false-positives (FP) the 956 ratings where benign lesions were actually cancer, false-negatives (FN) the 137 ratings where cancers were missed and true-positives (TP) the 1018 ratings where cancers were correctly detected (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Three patterns emerged: (i) extra redness, depression, fold deformation dominated correct cancer calls: they appeared in 76%, 70%, and 48% of TP ratings but in \u0026le;\u0026thinsp;21% of FN ratings, (ii) specificity loss was driven by \u0026ldquo;over-calling\u0026rdquo; redness and fold deformation\u0026mdash;both increased from 12\u0026ndash;17% in TN to about 40\u0026ndash;45% in FP decisions, (iii) rare yet vivid signs\u0026mdash;spontaneous bleeding and ulceration\u0026mdash;were almost absent in TN (\u0026le;\u0026thinsp;2.4%) but were observed significantly more in TP (24\u0026ndash;29%) (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003e\u0026ensp;Prevalence of each Blink cue within diagnostic quadrants (3300 post-intervention ratings)\u003c/b\u003e. Each cell shows the percentage of ratings in which the Blink feature was marked present. Columns correspond to the four combinations of ground-truth histology and participant prediction. CI: confidence interval.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue-negative\u003c/p\u003e \u003cp\u003e(benign \u0026harr; predicted benign)\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;1189\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFalse-positive\u003c/p\u003e \u003cp\u003e(benign \u0026harr; predicted cancer)\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;956\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFalse-negative\u003c/p\u003e \u003cp\u003e(cancer \u0026harr; predicted benign)\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;137\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTrue-positive\u003c/p\u003e \u003cp\u003e(cancer \u0026harr; predicted cancer)\u003c/p\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;1018\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eChicken-skin mucosa\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.5\u003c/p\u003e \u003cp\u003e(3.2\u0026ndash;6.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8.5\u003c/p\u003e \u003cp\u003e(4.7\u0026ndash;15.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e18.9\u003c/p\u003e \u003cp\u003e(14.4\u0026ndash;24.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18.9\u003c/p\u003e \u003cp\u003e(14.4\u0026ndash;24.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDepression\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9.3\u003c/p\u003e \u003cp\u003e(7.0\u0026ndash;12.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.9\u003c/p\u003e \u003cp\u003e(13.7\u0026ndash;30.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69.9\u003c/p\u003e \u003cp\u003e(62.7\u0026ndash;76.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e69.9\u003c/p\u003e \u003cp\u003e(62.7\u0026ndash;76.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFold deformation\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16.5\u003c/p\u003e \u003cp\u003e(13.0\u0026ndash;20.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.2\u003c/p\u003e \u003cp\u003e(13.2\u0026ndash;29.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47.9\u003c/p\u003e \u003cp\u003e(40.1\u0026ndash;55.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e47.9\u003c/p\u003e \u003cp\u003e(40.1\u0026ndash;55.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtra redness\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.3\u003c/p\u003e \u003cp\u003e(9.5\u0026ndash;15.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39.9\u003c/p\u003e \u003cp\u003e(29.5\u0026ndash;51.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e75.7\u003c/p\u003e \u003cp\u003e(69.2\u0026ndash;81.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e75.7\u003c/p\u003e \u003cp\u003e(69.2\u0026ndash;81.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSpontaneous bleeding\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.4\u003c/p\u003e \u003cp\u003e(1.6\u0026ndash;3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.8\u003c/p\u003e \u003cp\u003e(8.4\u0026ndash;21.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e29.3\u003c/p\u003e \u003cp\u003e(23.1\u0026ndash;36.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29.3\u003c/p\u003e \u003cp\u003e(23.1\u0026ndash;36.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eUlceration\u003c/b\u003e\u003c/p\u003e \u003cp\u003e% (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003cp\u003e(0.5\u0026ndash;1.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.9\u003c/p\u003e \u003cp\u003e(4.3\u0026ndash;14.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23.7\u003c/p\u003e \u003cp\u003e(18.3\u0026ndash;30.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e23.7\u003c/p\u003e \u003cp\u003e(18.3\u0026ndash;30.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eMultivariable modelling\u003c/p\u003e \u003cp\u003eThe mixed-effects logistic model confirmed three features as independent predictors of general endoscopists detecting cancer within a given polyp (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e): spontaneous bleeding: OR 3.92 (95%CI 3.11\u0026ndash;4.96), extra redness: OR 3.66 (3.09\u0026ndash;4.33) and depression: OR 3.06 (2.58\u0026ndash;3.64). Fold deformation trended towards significance (OR 1.18; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.060), whereas chicken-skin mucosa and ulceration were not significant after adjustment.\u003c/p\u003e \u003cp\u003eModel fit was excellent (AUC 0.79) (\u003cb\u003eSuppl. Table\u0026nbsp;8, Suppl. Figure\u0026nbsp;1\u003c/b\u003e), and the participant random effect collapsed to near-zero, indicating minimal unexplained between-rater heterogeneity once feature-level information was incorporated.\u003c/p\u003e \u003cp\u003eInter-observer agreement\u003c/p\u003e \u003cp\u003eRaw post-intervention agreement for presence/absence of each feature ranged from 0.53 (depression) to 0.94 (spontaneous bleeding). Fleiss\u0026rsquo; κ values were modest\u0026mdash;0.14 for fold deformation, 0.51 for spontaneous bleeding. Intraclass correlation coefficients for the composite Blink score varied from 0.45 to 0.75 across features (\u003cb\u003eSuppl Table\u0026nbsp;9\u003c/b\u003e).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis prospective, image-based interventional study demonstrates that the described 6 Blink features fundamentally address the performance gap of general endoscopists\u0026mdash;those who perform colonoscopy routinely but lack expertise in endoscopic imaging\u0026mdash;in detecting cancer within colorectal polyps. After a two-minute online introduction to the Blink-feature concept the cancer miss-rate fell four-fold. Importantly, the included images were taken using standard white-light endoscopy without magnification or virtual chromoendoscopy, making these improvements immediately accessible to the majority of endoscopists who lack advanced imaging expertise. The study's relevance lies not in enhancing expert performance\u0026mdash;specialists in endoscopic imaging have already developed sophisticated unconscious Blink pattern recognition\u0026mdash;but in democratizing cancer detection for the broader endoscopic community.\u003c/p\u003e\n\u003ch3\u003ePrimary endpoint – a large, uniform gain in sensitivity\u003c/h3\u003e\n\u003cp\u003eBreaking down the first-impression of cancer into six Blink features delivered a 21% absolute rise in sensitivity amongst general endoscopists. The effect was consistent across consultants, trainees, and surgeons. The magnitude of this effect\u0026mdash;reducing the miss rate from 27% to 6%\u0026mdash;represents a substantial improvement in this controlled setting. If similar gains in the live environment could be made may prevent up to one in five cancers from being inappropriately managed endoscopically. The price of this heightened vigilance was a 17% fall in specificity (from 75% to 58%). This trade-off is familiar in screening medicine: more disease is captured at the expense of more benign lesions being labelled suspicious[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe clinical implications of this sensitivity-specificity balance require careful consideration within the colorectal polyp management pathway. Over-calling benign polyps should trigger established safety nets\u0026mdash;multidisciplinary team review or referral for expert optical diagnosis\u0026mdash;that effectively mitigate potential harm[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. By contrast, the consequences of missing cancer are substantial and potentially irreversible: attempted pEMR with positive margins, perforation and tumor seeding, and, delay to definitive oncological therapy. Given these asymmetric consequences, the resultant sensitivity-specificity balance appears not only acceptable but potentially desirable for those endoscopists performing frontline assessments.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eSecondary endpoints\u003c/h2\u003e \u003cp\u003eTo understand why participants succeeded or failed after training, we mapped every rating onto a 2\u0026times;2 matrix of ground-truth histology (benign/cancer) versus participant verdict (benign/cancer). Three distinct patterns emerged from this analysis:\u003c/p\u003e \u003cp\u003e \u003cem\u003e1. Blink features that secure correct cancer calls\u003c/em\u003e \u003c/p\u003e \u003cp\u003eWhen cancers were correctly identified, extra redness, depression, and fold deformation were the commonest cues recognized by participants. These same features showed markedly different prevalence in false-negative decisions, suggesting that failure to recognize these signs\u0026mdash;rather than their absence\u0026mdash;drove most misclassifications. This pattern implies that further refinement of teaching materials could focus on threshold recognition: at what point does mild erythema become \"extra redness,\" and when does surface irregularity constitute a true \"depression\".\u003c/p\u003e \u003cp\u003e2. Features that erode specificity\u003c/p\u003e \u003cp\u003eFalse-positive escalation of benign polyps was largely driven by over-calling extra redness and fold deformation. The high false-positive rate for fold deformation may reflect a specific clinical scenario: scarring from previous incomplete resection can produce architectural distortion mimicking invasive growth. Teaching must therefore refine when these two signs truly imply invasion versus benign fibrosis, potentially by including number of folds converging within the polyp or combining with high specificity features. Chicken-skin mucosa added little independent information once the other cues were known in this dataset, suggesting potential redundancy in the current six-item framework\u003c/p\u003e \u003cp\u003e3. Rare but powerful signs\u003c/p\u003e \u003cp\u003eSpontaneous bleeding and ulceration remained almost absent in true-negatives yet increased dramatically in true-positives, explaining their very high specificities and strong adjusted odds ratios in the multivariate model. These \"high-value\" signs function as near-pathognomonic markers of invasion when present, though their relative rarity in the dataset limited their contribution to overall sensitivity.\u003c/p\u003e \u003cp\u003eTogether these patterns both help us understand how cancer decisions are made by general endoscopists and hint at a targeted refinement for the next training iteration: raise awareness that there should be a higher diagnostic threshold/information for extra redness and fold deformation while reinforcing that spontaneous bleeding and ulceration, though uncommon, are high-yield indicators of malignancy. Seen differently these are precisely the situations where expert assessment is important; these features can identify areas of polyps which can be carefully interrogated with magnification and chromoendoscopy for pit/vascular pattern analysis to reach a precise diagnosis. However, the lesion must first be detected and escalated to the expert and so both strategies and structured communication between colleagues are therefore essential.\u003c/p\u003e \u003cp\u003eThe multivariate model's random-participant variance collapsed towards zero, indicating that once feature use was accounted for, little unexplained heterogeneity remained between endoscopists. Such a pattern strongly supports this structured training approach over experiential learning alone.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInter-observer agreement\u003c/h3\u003e\n\u003cp\u003eDespite increased sensitivity using the framework, Fleiss' κ for most features stayed in the slight-to-fair range, with only spontaneous bleeding achieving moderate agreement. This persistent inter-observer variability is not unique to our study[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Importantly, however, despite modest feature-level agreement the overall framework still produced substantial improvements in overall diagnostic performance, suggesting that its value lies in its overall outcome (number of features) rather than perfect concordance on individual features.\u003c/p\u003e\n\u003ch3\u003eExperience still helps, but structure helps more\u003c/h3\u003e\n\u003cp\u003e Amongst participants, those with extensive colonoscopy experience posted the highest post-training sensitivity, yet trainees showed the greatest absolute gain, underscoring that explicit deconstruction of a problem can compress learning curves. This differential improvement pattern reveals an important distinction: experience in performing colonoscopy does not automatically confer expertise in endoscopic imaging. Many experienced gastroenterologists lack formal training in optical diagnosis, while conversely, some junior endoscopists may have received structured imaging education. The Blink framework bridges this specific skill gap by making a certain amount of tacit imaging knowledge explicit, allowing general endoscopists to appropriate pattern recognition strategies typically reserved for imaging specialists.\u003c/p\u003e \u003cp\u003eThat said, specificity dropped equally across all experience strata, reinforcing that if considering improving upon these results in the future the challenge lies not in experience \u003cem\u003eper se\u003c/em\u003e but in calibrating appropriate thresholds for suspicious features and criteria for expert referral via multi-disciplinary discussion.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eGrounding the findings in learning theory\u003c/h2\u003e \u003cp\u003eAccording to dual-process models of cognition[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], experts use fast, pattern-recognition \"System 1\" heuristics, while novices rely on slower, analytic \"System 2\" reasoning. Our data suggest that explicitly labelling key visual patterns (introducing Blink features framework) allows general endoscopists to appropriate elements of System 1 thinking almost immediately\u0026mdash;a concept validated in parallel fields of medicine [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. First, it reduces intrinsic cognitive load by chunking complex visual information into six discrete, memorable features. Second, it minimizes extraneous load by providing a consistent evaluation sequence. Third, it converts vague intuitions (\"this looks worrying\") into explicit, teachable schemas that can be rehearsed and automated through deliberate practice. This scaffolding is particularly valuable in endoscopy, where real-time decision-making under the cognitive pressure of a complex motor skill often precludes extensive deliberation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eStrengths and limitations\u003c/h2\u003e \u003cp\u003eStrengths of this study include the large globally diverse cohort, a consensual expert gold standard with forced consensus methodology, pre-post design and mixed-effects statistics that appropriately account for clustering within participants. The image quality\u0026mdash;obtained from standard-definition endoscopes with good preparation\u0026mdash;likely represents real-life conditions for visualization.\u003c/p\u003e \u003cp\u003eSeveral limitations warrant mention. First, our 20-image set was enriched for cancer (~\u0026thinsp;35%), potentially boosting sensitivity relative to everyday practice; repeating the identical images after the learning-intervention also risks a recall effect (somewhat mitigated by randomization). Second, the small sample restricts statistical power and makes κ estimates imprecise. Third, still photographs omit dynamic cues (bleeding on contact, fold deformation during peristalsis) and the cognitive load of live colonoscopy, so translation to real-time performance remains unproven. Fourth, the study measures diagnostic accuracy only; we did not track downstream outcomes to show fewer interval cancers or avoided surgeries, nor did we quantify the additional referrals generated by the specificity drop. Fifth, voluntary, education-minded participants may overstate uptake compared with the broader endoscopy workforce. Finally, our binary benign-vs-cancer framework does not accommodate the clinical grey zone of indeterminate lesions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFuture directions: \u0026ldquo;Blink first, then look closer\u0026rdquo;\u003c/h2\u003e \u003cp\u003eWe propose a two-tier implementation pathway integrating rapid screening with detailed analysis. When an LNPCP is identified, the operator pauses the endoscopic image and performs a rapid assessment of the six Blink features. If two or more are present, this triggers mandatory escalation. In interventional endoscopy settings, this means proceeding to (virtual) chromoendoscopy and magnification to evaluate established criteria (e.g. vascular/pit pattern analysis). In screening examinations, high-quality images and video should be captured for multidisciplinary team review. This approach respects efficiency demands while providing a safety net against the documented false-positive tendency.\u003c/p\u003e \u003cp\u003eFor those seeking to optimize their specificity using this framework, threshold refinement through borderline case libraries\u0026mdash;particularly for redness and fold deformation\u0026mdash;may further enhance performance. However, the core value proposition remains sensitivity improvement for general endoscopists. The six Blink features also constitute human-interpretable labels that could seed explainable AI algorithms, enabling real-time decision support that transparently indicates which features triggered concern.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eA concise six-item Blink framework reduced the miss rate for invasive cancer in LNPCP still-images amongst general endoscopists 4-fold, while revealing precisely which visual cues drive both successful detection and false alarms. Embedding this structured approach into training curricula, coupled with selective use of confirmatory imaging for screen-positive lesions could offer an evidence-based pathway to reduce missed cancers while managing the increased false-positive rate through existing clinical safeguards. Further validation in live endoscopy settings will determine whether these gains in still-image interpretation translate to meaningful improvements in real-world patient outcomes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003ePotential competing interests and funding statement \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical.\u003cbr\u003e\u0026nbsp;The remaining authors have no competing interests to disclose.\u003c/p\u003e\n\u003cp\u003eThe University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript.\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003c/p\u003e\u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003eThe study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of Ghent University Hospital (ONZ-2022-0488). The study was registered on ClinicalTrials.gov (NCT05699954) on 20 April 2023. Informed consent was obtained from participating patients and endoscopists prior to the study.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable. No individual participant data or identifiable images are included in this manuscript.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e\u003ch2\u003eCompeting Interests\u003c/h2\u003e\u003cp\u003eDJT: consulting fees and research support from Olympus Medical, Fujifilm Europe and Pentax Medical.The remaining authors have no competing interests to disclose.The University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe University Hospital of Ghent (UZ Gent), Belgium provided funding for a research nurse to assist with the administration of the study. There was no influence from the institution regarding study design or conduct, data collection, management, analysis, or interpretation or preparation, review, or approval of the manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eLynn Debels Analysed data, wrote the manuscript and revised according to author comments Tamas Tornai Analysed data, assisted with writing and revision of manuscriptJohn Anderson Assisted with writing and revision of manuscript Maria Eva Argenziano Assisted with writing and revision of manuscript Anne Hoorens Histopathological analysisVikash Lala Assisted with writing and revision of manuscript Pieter Jan Poortmans Assisted with writing and revision of manuscript Roland Valori Assisted with writing and revision of manuscript Lobke Desomer Assisted with writing and revision of manuscript David J Tate Designed study, designed algorithm, performed procedures, collected data, analyzed data, wrote manuscript, revised manuscript after co-authors remarks\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e \u003cp\u003eNot applicable\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eZauber AG, Winawer SJ, O'Brien MJ, et al. Colonoscopic polypectomy and long-term prevention of colorectal-cancer deaths. N Engl J Med. 2012;366:687\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoss A, Bourke MJ, Williams SJ, et al. Endoscopic mucosal resection outcomes and prediction of submucosal cancer from advanced colonic mucosal neoplasia. Gastroenterology. 2011;140:1909\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1053/j.gastro.2011.02.062\u003c/span\u003e\u003cspan address=\"10.1053/j.gastro.2011.02.062\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJayanna M, Burgess NG, Singh R, et al. Cost Analysis of Endoscopic Mucosal Resection vs Surgery for Large Laterally Spreading Colorectal Lesions. Clin Gastroenterol hepatology: official Clin Pract J Am Gastroenterological Association. 2016;14:271\u0026ndash;2.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeswani RN, Law R, Ciolino JD, et al. Adverse events after surgery for nonmalignant colon polyps are common and associated with increased length of stay and costs. Gastrointest Endosc. 2016;84:296\u0026ndash;e303291.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaito Y, Sakamoto T, Dekker E, et al. First report from the International Evaluation of Endoscopic classification Japan NBI Expert Team: International multicenter web trial. Dig Endosc. 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/den.14682\u003c/span\u003e\u003cspan address=\"10.1111/den.14682\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDekker E, Houwen B, Puig I, et al. Curriculum for optical diagnosis training in Europe: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy. 2020;52:899\u0026ndash;923. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1055/a-1231-5123\u003c/span\u003e\u003cspan address=\"10.1055/a-1231-5123\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeulen LWT, van de Wetering AJP, Debeuf MPH, et al. Optical diagnosis of T1 CRCs and treatment consequences in the Dutch CRC screening programme. Gut. 2020;69:2049\u0026ndash;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/gutjnl-2019-320403\u003c/span\u003e\u003cspan address=\"10.1136/gutjnl-2019-320403\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeulen LWT, Haasnoot KJC, Vlug MS, et al. Effect of optical diagnosis training on recognition and treatment of submucosal invasive colorectal cancer in community hospitals: a prospective multicenter intervention study. Endoscopy. 2024;56:770\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1055/a-2313-4996\u003c/span\u003e\u003cspan address=\"10.1055/a-2313-4996\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKahneman Da. Thinking, fast and slow. 1st ed. New York: Farrar, Straus and Giroux, [2011] \u0026copy;2011; 2011.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaat EM, Farr I, Wolfe JM, et al. Comparable prediction of breast cancer risk from a glimpse or a first impression of a mammogram. Cogn Res Princ Implic. 2021;6:72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s41235-021-00339-5\u003c/span\u003e\u003cspan address=\"10.1186/s41235-021-00339-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEvans KK, Georgian-Smith D, Tambouret R, et al. The gist of the abnormal: above-chance medical decision making in the blink of an eye. Psychon Bull Rev. 2013;20:1170\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3758/s13423-013-0459-3\u003c/span\u003e\u003cspan address=\"10.3758/s13423-013-0459-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePuig I, M\u0026aacute;rmol C, Bustamante M. Endoscopic imaging techniques for detecting early colorectal cancer. Curr Opin Gastroenterol. 2019;35:432\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/mog.0000000000000570\u003c/span\u003e\u003cspan address=\"10.1097/mog.0000000000000570\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBugajski M, Kaminski MF, Orlowska J, et al. Suspicious macroscopic features of small malignant colorectal polyps. Scand J Gastroenterol. 2015;50:1261\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3109/00365521.2015.1024280\u003c/span\u003e\u003cspan address=\"10.3109/00365521.2015.1024280\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuan J, Zhao R, Zhang X, et al. Chicken skin mucosa surrounding adult colorectal adenomas is a risk factor for carcinogenesis. Am J Clin Oncol. 2012;35:527\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/COC.0b013e31821dedf7\u003c/span\u003e\u003cspan address=\"10.1097/COC.0b013e31821dedf7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHorie H, Togashi K, Kawamura YJ, et al. Colonoscopic stigmata of 1 mm or deeper submucosal invasion in colorectal cancer. Dis Colon Rectum. 2008;51:1529\u0026ndash;34. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10350-008-9263-y\u003c/span\u003e\u003cspan address=\"10.1007/s10350-008-9263-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIkehara H, Saito Y, Matsuda T, et al. Diagnosis of depth of invasion for early colorectal cancer using magnifying colonoscopy. J Gastroenterol Hepatol. 2010;25:905\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.1440-1746.2010.06275.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1440-1746.2010.06275.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJang HW, Park SJ, Cheon JH, et al. Does magnifying narrow-band imaging or magnifying chromoendoscopy help experienced endoscopists assess invasion depth of large sessile and flat polyps? Dig Dis Sci. 2014;59:1520\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10620-014-3090-x\u003c/span\u003e\u003cspan address=\"10.1007/s10620-014-3090-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKudo S, Hirota S, Nakajima T, et al. Colorectal tumours and pit pattern. J Clin Pathol. 1994;47:880\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/jcp.47.10.880\u003c/span\u003e\u003cspan address=\"10.1136/jcp.47.10.880\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKudo S, Kashida H, Tamura T, et al. Colonoscopic diagnosis and management of nonpolypoid early colorectal cancer. World J Surg. 2000;24:1081\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s002680010154\u003c/span\u003e\u003cspan address=\"10.1007/s002680010154\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee YM, Song KH, Koo HS, et al. Colonic Chicken Skin Mucosa Surrounding Colon Polyps Is an Endoscopic Predictive Marker for Colonic Neoplastic Polyps. Gut Liver. 2022;16:754\u0026ndash;63. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5009/gnl210271\u003c/span\u003e\u003cspan address=\"10.5009/gnl210271\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatsuda T, Saito Y, Hotta K, et al. Prevalence and clinicopathological features of nonpolypoid colorectal neoplasms: should we pay more attention to identifying flat and depressed lesions? Dig Endosc. 2010;22(Suppl 1):S57\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.1443-1661.2010.00967.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1443-1661.2010.00967.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMisawa M, Kudo SE, Wada Y, et al. Magnifying narrow-band imaging of surface patterns for diagnosing colorectal cancer. Oncol Rep. 2013;30:350\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3892/or.2013.2471\u003c/span\u003e\u003cspan address=\"10.3892/or.2013.2471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNowicki MJ, Bishop PR, Subramony C, et al. Colonic chicken-skin mucosa in children with polyps is not a preneoplastic lesion. J Pediatr Gastroenterol Nutr. 2005;41:600\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/01.mpg.0000179658.09210.b6\u003c/span\u003e\u003cspan address=\"10.1097/01.mpg.0000179658.09210.b6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePuig I, L\u0026oacute;pez-Cer\u0026oacute;n M, Arnau A, et al. Accuracy of the Narrow-Band Imaging International Colorectal Endoscopic Classification System in Identification of Deep Invasion in Colorectal Polyps. Gastroenterology. 2019;156:75\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1053/j.gastro.2018.10.004\u003c/span\u003e\u003cspan address=\"10.1053/j.gastro.2018.10.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaito Y, Fujii T, Kondo H, et al. Endoscopic treatment for laterally spreading tumors in the colon. Endoscopy. 2001;33:682\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1055/s-2001-16213\u003c/span\u003e\u003cspan address=\"10.1055/s-2001-16213\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaitoh Y, Obara T, Watari J, et al. Invasion depth diagnosis of depressed type early colorectal cancers by combined use of videoendoscopy and chromoendoscopy. Gastrointest Endosc. 1998;48:362\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s0016-5107(98)70004-5\u003c/span\u003e\u003cspan address=\"10.1016/s0016-5107(98)70004-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShatz BA, Weinstock LB, Thyssen EP, et al. Colonic chicken skin mucosa: an endoscopic and histological abnormality adjacent to colonic neoplasms. Am J Gastroenterol. 1998;93:623\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.1572-0241.1998.177_b.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1572-0241.1998.177_b.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUno Y, Munakata A. Endoscopic and histologic correlates of colorectal polyp bleeding. Gastrointest Endosc. 1995;41:460\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s0016-5107(05)80004-5\u003c/span\u003e\u003cspan address=\"10.1016/s0016-5107(05)80004-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTate DJ. Introduction to 6 Blink Features - Learning Tool. In: Online: 2022: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://vimeo.com/768168090\u003c/span\u003e\u003cspan address=\"https://vimeo.com/768168090\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBates D, M\u0026auml;chler M, Bolker B, et al. Fitting Linear Mixed-Effects Models Using lme4. J Stat Softw. 2015;67:1\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.18637/jss.v067.i01\u003c/span\u003e\u003cspan address=\"10.18637/jss.v067.i01\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eemmeans RL. Estimated Marginal Means, aka Least-Squares Means. R package version 1.11.2-00002,. 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaughnessy AF. Clinical Epidemiology: A Basic Science for Clinical Medicine. BMJ. 2007;335:777.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Fabio F, Jitsumura M, Longstaff L, et al. Management of Significant Polyp and Early Colorectal Cancer Is Optimized by Implementation of a Dedicated Multidisciplinary Team Meeting: Lessons Learned From the United Kingdom National Program. Dis Colon Rectum. 2022;65:654\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/dcr.0000000000002199\u003c/span\u003e\u003cspan address=\"10.1097/dcr.0000000000002199\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Doorn SC, Hazewinkel Y, East JE, et al. Polyp morphology: an interobserver evaluation for the Paris classification among international experts. Am J Gastroenterol. 2015;110:180\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/ajg.2014.326\u003c/span\u003e\u003cspan address=\"10.1038/ajg.2014.326\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8553404/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8553404/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground \u0026amp; Aims:\u003c/h2\u003e \u003cp\u003eAccurate cancer detection in large non-pedunculated colorectal polyps (LNPCPs) remains challenging for general endoscopists. We evaluated whether teaching six gross morphological \"Blink\" features could improve the accuracy of cancer detection.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis prospective interventional study assessed general endoscopists evaluating 20 LNPCP images (7 with histologically-confirmed submucosal invasive cancer including 4 deep invasions\u0026thinsp;\u0026ge;\u0026thinsp;1000\u0026micro;m, 13 benign). Participants assessed images before and after a 2-minute educational video introducing six Blink features: spontaneous bleeding, depression, fold deformation, extra redness, ulceration, and chicken-skin mucosa. Primary outcome was change in miss-rate for cancer detection. Generalized linear mixed models accounted for clustering within raters and polyps.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003e165 participants included gastroenterology consultants (63.6%), trainees (21.2%), students (1.8%) and colorectal surgeons (13.3%) with a median colonoscopy experience of 6.5 years. Post-intervention the cancer miss-rate decreased four-fold from 26.6% (95%CI 13.4\u0026ndash;46.0) to 5.7% (95%CI 2.4\u0026ndash;12.8). This improvement was consistent across experience levels. The false alarm rate increased less-than two-fold from 25.0% (95%CI 15.1\u0026ndash;38.5) to 42.2% (95%CI 27.7\u0026ndash;58.2). Multivariable analysis identified spontaneous bleeding (OR 3.92, 95%CI 3.11\u0026ndash;4.96), extra redness (OR 3.66, 95%CI 3.09\u0026ndash;4.33), and depression (OR 3.06, 95%CI 2.58\u0026ndash;3.64) as independent predictors of cancer amongst general endoscopists. Mean Blink features per polyp were 1.08 (95%CI 0.82\u0026ndash;1.40) for benign lesions versus 2.46 (95%CI 1.85\u0026ndash;3.12) for cancers.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eTeaching six Blink features to general endoscopists led to a 4x reduction in cancer miss-rates in an image-based evaluation. While specificity decreased, this trade-off favors patient safety, as false-positives trigger established clinical safeguards while missed cancers risk inappropriate endoscopic resection with potentially irreversible consequences.\u003c/p\u003e\u003ch2\u003eGraphical Abstract\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e","manuscriptTitle":"Teaching Six ‘Blink’ Features Reduces General Endoscopists’ Cancer Miss-Rate in Image-Based Assessment of Large Colorectal Polyps","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-17 14:46:28","doi":"10.21203/rs.3.rs-8553404/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d515c868-4375-467b-9285-e63ef3a8355c","owner":[],"postedDate":"February 17th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-18T09:28:09+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-17 14:46:28","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8553404","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8553404","identity":"rs-8553404","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}