Diagnostic performance and agreement of auditors for evaluation of computer-aided optical polyp diagnosis: Prospective study

Abstract

Background and study aims

Guidelines recommend independent auditing of diagnostic performance for clinical implementation of computer-aided optical polyp diagnosis (CADx). This study evaluated diagnostic performance and interobserver agreement of auditors and offered guidance on conducting CADx audits.

Methods

Images and videos of all ≤ 5-mm polyps from a large, prospective study with systematic activation of CADx were audited by three expert endoscopists. Experts performed independent, blinded diagnostic review including documentation of confidence level. The primary outcome was sensitivity of audit by three experts for high-confidence adenomas compared with pathology. Secondary outcomes included number of reviewers for optimal CADx auditing and interobserver agreement.

Results

Four hundred eighty-seven diminutive polyps were audited (510 patients). Sensitivity was 99.4% (95% confidence interval [CI] 96.0–100) using three experts (Strategy A); 88.7% (95% CI 84.1–92.1) using two experts and one referee (Strategy B); 99% (95% CI 96–99.8), 98.8% (95% CI 95.4–99.8), and 99.4% (95% CI 96.3–100) using two-expert combinations (Strategy C); and 98.2% (95% CI 95.1–99.4), 97.3% (95% CI 94.0–98.9), and 88.9% (95% CI 83.6–92.7) for each expert individually (Strategy D). Among 266 pathology-based adenomas, Strategy A evaluated 160 polyps versus 196, 172, and 170 in Strategy C; and 220, 223, and 207 in Strategy D. Strategy B evaluated all 266 adenomas. Overall interobserver agreement was moderate (kappa 0.52), but very high for high-confidence adenomas (kappa 0.89).

Conclusions

Expert audit for evaluating CADx resulted in high sensitivity and interobserver agreement for high-confidence adenomas. Audit by two experts, with a third expert for arbitration, permitted audit of all polyps and effective assessment of CADx within clinical studies.

 

Scroll to top

Keywords

Endoscopy Lower GI Tract - Polyps / adenomas / ... - GI Pathology - Quality and logistical aspects - Quality management

Introduction

Histopathology is currently considered the gold standard for diagnostic evaluation of diminutive (≤ 5 mm) colorectal polyps. Clinical implementation of real-time optical polyp diagnosis using computer-aided diagnosis systems (CADx) would allow histopathology to be replaced, reducing resource utilization and healthcare costs [1] [2]. CADx-assisted optical diagnosis for real-time prediction of polyp histology might enable two key optical diagnosis-based management strategies: 1) “resect and discard,” where diminutive polyps deemed neoplastic are excised and discarded without histopathological assessment; and 2) “diagnose and leave,” where hyperplastic lesions in the rectosigmoid are left in situ following optical diagnosis [3].

Several large studies have shown adequate diagnostic performance for CADx-assisted and CADx-unassisted optical diagnosis [4] [5] [6]. A recent study demonstrated the pragmatic implementation of resect and discard and diagnose and leave strategies using CADx-assisted optical diagnosis [7]. However, when implementing such optical diagnosis-based strategies, an independent video- or image-based audit of diagnostic performance becomes essential. European Society of Gastrointestinal Endoscopy guidelines recommend audit of polyps undergoing optical diagnosis [8]. However, methodology for conducting such audits and associated diagnostic accuracy compared with pathology remain unclear.

To address this knowledge gap, we evaluated audit performance in a prospective video-based study using video and image material from a large prospective study with systematic CADx-assisted optical diagnosis and video recording of all procedures. The study aim was to provide information on performance of diagnostic audit and offer a framework for how audit of CADx-assisted optical diagnosis should be performed.

 

Scroll to top

Patients and methods

This study is reported according to the Standards for Reporting of Diagnostic Accuracy Studies (STARD) checklist [9]. All authors had access to the study data and reviewed and approved the final manuscript.

Study design and participants

This study was a prospective study using data from a previous prospective trial evaluating CADx-assisted optical polyp diagnoses [10]. Consecutive patients aged 45 to 80 years were enrolled at a single center (Montreal University Hospital Center) and underwent elective colonoscopy with CADx-assisted optical diagnosis. Consent for study participation and secondary use of data was obtained from all patients. Secondary data use and the audit evaluation subproject were preconceived in the original protocol and approved by the Institutional Review Board (CER 21.305; NCT05236790). Patient inclusion and exclusion criteria have been reported previously [10].

 

Scroll to top

Study procedures

High-definition endoscopes (ELUXEO 7000 System, EC-760S-A/M and EC-760S-A/L colonoscopes; Fujifilm, Tokyo, Japan) were used for colonoscopies. The CAD-EYE software (EW10-EC02; Fujifilm) was used for imaging of all diminutive colorectal polyps (≤ 5 mm). Four board-certified gastroenterologists or one trainee performed all colonoscopies. The linked-color and blue-light imaging functions were activated to enhance visualization of polyp microscopic features. Polyp size was measured optically by the endoscopist or using the CAD-EYE integrated virtual eye scale (Scale-eye). Polyps were assessed using CADx-assisted optical diagnosis and full-length video recordings of the procedures were retained. All polyps were resected and sent to pathology for review by a single board-certified pathologist who specialized in gastrointestinal diseases. Polyps were cut into at least three levels and classified according to the World Health Organization classification system [11].

In the current study, image and video sequences of each diminutive polyp were extracted from the colonoscopy recordings by research assistants (F.H., T.I.D., D.H.D.K, B.A., N.H., S.F.). All retrieved diminutive polyps from the index study were used for inclusion in this secondary analysis. The materials were incorporated into an online questionnaire using LimeSurvey (Hamburg, Germany), with each page exhibiting a different polyp. The survey was used for audit by three independent expert endoscopists (D.K.R., H.P., C.S.). Each expert had more than 10 years of endoscopy experience and participated in at least 10 prior optical diagnosis and CADx studies. Research assistants (F.H., T.I.D., D.H.D.K.) helped the experts navigate the survey. Experts were provided with images and short video sequences showing each polyp along with information on anatomical location and polyp size but were blinded to each other's diagnosis, CADx optical diagnosis, and pathology results. Experts classified each polyp as adenoma, hyperplastic, sessile serrated lesion (SSL), normal/mucosal prolapse, inflammatory, or unclear. They rated their confidence in their diagnosis as high or low. The experts also indicated whether overall quality of the image and video for each polyp was adequate or inadequate for audit. Quality documentation was determined by the “3 C” endoscopy photo quality checklist proposed by Ahmad et al. (clean mucosal surface, complete view, correct focal distance), as well as suitability of image modalities (white-light imaging, linked-color imaging, blue-light imaging) for polyp visualization [4].

 

Scroll to top

Data collection and outcomes

Collected data included patient age, sex, American Society of Anesthesiologists class, Boston Bowel Preparation Scale score, and colonoscopy indication; polyp location, size, and histology; experts’ optical diagnoses, levels of confidence, and ratings of image and video quality. The primary outcome of the study was sensitivity of a three-expert review with high confidence, using histopathology assessment as the reference diagnosis. We hypothesized that sensitivity of histology prediction by experts would be superior to the 80% threshold defined by the ESGE Simple Optical Diagnosis Accuracy competence standards [12]. Secondary exploratory outcomes were: 1) to determine the optimal audit strategy for CADx-assisted optical diagnosis; and 2) expert interobserver agreement.

Audit strategy choices were:

• Strategy A - Review by three experts with concordant, high-confidence diagnoses;

• Strategy B - review by the first two experts (D.K.R., H.P.) with concordant, high-confidence diagnoses, with a referee (C.S.) in cases of disagreement;

• Strategy C -Two-expert audit, yielding three possible pairs of experts, because three experts participated in the study;

• Strategy D - Audit performed by each expert individually.

All strategies were evaluated and compared with each other using performance data (accuracy, sensitivity, specificity) and polyp sample sizes. Accuracy was defined as the number of polyps correctly diagnosed by the experts, all types combined, divided by the total number of polyps. Sensitivity was defined as the proportion of adenomas correctly diagnosed, and specificity was defined as the proportion of non-adenomatous polyps correctly identified as such.

 

Scroll to top

Sample size and statistical analysis

We assumed 90% sensitivity of expert audit when performed by three experts with concordant, high-confidence diagnoses. Therefore, to conduct a two-sided chi-squared test with a significance level of 5% and power of 80% to test whether the experts’ sensitivity was greater than 80%, it was necessary to include a minimum of 98 polyps with adenoma histology and concordant, high-confidence diagnosis by all experts.

Descriptive statistics are presented as numbers and frequencies for categorical variables, and as means with standard deviation (SD) for continuous variables. Diagnostic performance metrics were calculated for different combinations of experts with high-confidence diagnoses (i.e. for all three experts agreeing on the diagnosis (Strategy A), for two experts with a referee in cases of disagreement (Strategy B), for two experts agreeing on the diagnosis (Strategy C), and for each expert individually (Strategy D). For all metrics (accuracy, sensitivity, specificity), generalized estimating equations assuming exchangeable intra-patient correlation between the polyps led to an estimated correlation close to zero; hence, independence was assumed. Estimated overall accuracy, adenoma sensitivity and specificity are presented with 95% confidence intervals (CIs), and two-tailed chi-squared tests with a 5% significance level were considered significant for all analyses. Light’s kappa statistics were used to determine interobserver agreements. Analyses were performed using R 4.4.2 (R Core Team, 2024; R Foundation for Statistical Computing, Vienna, Austria).

 

Scroll to top

 

Scroll to top

Results

Patient and polyp characteristics

Among 510 patients, 523 diminutive (≤ 5 mm) colorectal polyps were identified, of which 36 polyps were excluded from the audit (18 because they were visible alongside another polyp, 16 because video recordings were missing, 1 because pathology results were not available, and 1 because of suboptimal video quality) ([Fig. 1]). Finally, 487 polyps were independently reviewed by each of the three expert endoscopists, of which 266 (54.6%) were histologically diagnosed as adenomas, 130 (26.7%) as hyperplastic polyps, 16 (3.3%) as SSLs, and 63 (12.9%) as normal mucosa ([Table 1]). Of the 266 adenomas, the three-expert audit reached concordant, high-confidence diagnosis for 160 polyps (60.2%). Two experts reported using images more frequently than videos for polyp diagnosis throughout the review, and one reported predominant use of videos. Pathology reported high-grade dysplasia in two of 487 reviewed polyps (0.4%), which went undetected by the experts.

Fig. 1 Study flow chart.

Table 1 Patient and polyp characteristics.

Characteristics
HGD, high-grade dysplasia; SSL, sessile serrated polyp.
Patients, n	510
Age, mean (SD), years	64.0 (8.6)
Female, n (%)	254 (49.8)
Colonoscopy indication, n (%)
Screening	72 (14.1)
Surveillance	274 (53.7)
Diagnostic	142 (27.8)
Polypectomy	16 (3.1)
Polyps included in expert audit, n	487
Size, mm, n (%)
1–3	333 (68.4)
3–5	154 (31.6)
Polyp location, n (%)
Cecum	50 (10.3)
Ascending	110 (22.6)
Hepatic flexure	2 (0.4)
Transverse	110 (22.6)
Descending	81(16.6)
Sigmoid	98 (20.1)
Rectum	36 (7.4)
Histology, n (%)
Adenoma	266 (54.6)
Hyperplastic polyp	130 (26.7)
SSL	16 (3.3)
Normal mucosa	63 (12.9)
Inflammatory polyp	8 (1.6)
Other	4 (0.8)
HGD, n (%)	2 (0.4)

 

Scroll to top

Three-expert review sensitivity

Audit of adenomas by three experts with concordant, high-confidence diagnosis (Strategy A) yielded 99.4% sensitivity (159/160, 95%CI 96.0–100) ([Fig. 2]). This sensitivity was found to be significantly higher than the 80% threshold defined by the ESGE guidelines (P < 0.001).

Fig. 2 Diagnostic characteristics of each audit strategy (%, 95% confidence interval [CI]). For each performance metric and audit strategy, the number of polyps included in the denominator is indicated at the bottom of the panels.

 

Scroll to top

Choice of audit strategy

Overall expert accuracy was 76.9% (176/229, 95% CI 70.7–82.0) in Strategy A, whereas it was 63.2% (308/487, 95% CI 58.8–67.5) in a review by two experts and a referee (Strategy B) ([Fig. 2]). The three possible pairs of experts in Strategy C had similar accuracies (75.1% [95% CI 69.6–79.9], 77% [95% CI 71.2–81.9], and 74.8% [95% CI 69.1–79.8]). In Strategy D, each expert individually achieved different levels of accuracy (70.5% [95% CI 65.4–75.2], 71.9% [95% CI 66.9–76.3], and 65.3% [95% CI 60.3–70.1]).

Sensitivity was 88.7% (236/266, 95%CI 84.1–92.1) in Strategy B, whereas sensitivities in Strategy C were 99.0% (194/196, 95%CI 96.0–99.8), 98.8% (170/172, 95%CI 95.4–99.8), and 99.4% (169/170, 95%CI 96.3–100). Strategies A and C thus yielded similar sensitivities with overlapping 95%CIs ([Fig. 2]). Different sensitivities were obtained in Strategy D for each expert (98.2% [95% CI 95.1–99.4], 97.3% [95% CI 94.0–98.9], and 88.9% [95% CI 83.6–92.7]).

Specificity was 40.6% (95%CI 29.1–53.1) in Strategy A and was higher in Strategy B (63.8% [95% CI 57.0–70.1]). Specificities in Strategy C varied depending on the pair of experts (37.1% [95%CI 27.3–48.0], 48.8% [95%CI 37.5–60.1], and 51% [95%CI 40.7–61.3]). Specificities also varied significantly in Strategy D (42.9% [95% CI 34.2–52], 51% [95% CI 42.6–59.4], 69% [61.4–75.7]).

For each strategy, the number of experts influenced the number of polyps with concordant, high-confidence diagnoses that were included in performance assessments. Among all polyp types, Strategy B allowed inclusion of all polyps (N = 487), compared with 229 in Strategy A; 285, 252, and 266 in Strategy C; and 346, 366, and 378 in Strategy D. When calculating sensitivities and therefore only considering all adenomas (N = 266), Strategy A allowed inclusion of 160 polyps, versus 196, 172, and 170 polyps in Strategy C; and 220, 223, and 207 in Strategy D. In Strategy B, all 266 adenomas were included.

 

Scroll to top

Interobserver agreement

Overall agreement between the three experts was moderate for all polyps, with a kappa value of 0.52, and substantial for adenomas (kappa 0.66) ([Fig. 3]). Interobserver agreement for high-confidence diagnoses was very high, with a kappa value of 0.83 overall. More specifically, consensus was the highest for high-confidence diagnoses of adenomas (kappa 0.89), whereas it was lower for hyperplastic polyps (kappa 0.82) and even lower for SSLs (kappa 0.69).

Fig. 3 Expert interobserver agreement according to polyp histology and confidence level. Dark blue: no agreement; dark green: near perfect agreement.

 

Scroll to top

 

Scroll to top

Discussion

In this prospective study, we found that sensitivity of expert review was very high for ≤ 5-mm polyps unanimously diagnosed as high-confidence adenomas. Regardless of the number of auditors, expert performance in diagnosing adenomas was found to be likely sufficient for audit of CADx-assisted optical diagnosis. Furthermore, our findings suggest that the ideal audit strategy is an initial high-confidence review by two blinded experts, with a third expert available for arbitration in cases of disagreement. Although this approach yielded lower accuracy and sensitivity compared with other strategies (three-expert audit, two-expert audits, single-expert audit), Strategy B allowed all polyps to be included, making it a more practical approach for auditing CADx. Moreover, specificity of Strategy B was higher than that of Strategies A and C, and was less affected by individual expert performances such as in Strategy D. Nevertheless, suboptimal specificity across all audit strategies highlights the tendency of endoscopists to over-diagnose adenomas, which has also been reported previously [13]. A significant proportion of pathology-based hyperplastic polyps or normal mucosa have likely been diagnosed as high-confidence adenomas by the experts. This phenomenon could be due to endoscopists’ fear of misdiagnosing adenomas as another benign lesion because of patient safety concerns. However, inaccurate histological diagnosis of high-confidence adenomas as normal mucosa due to errors in polyp retrieval and sectioning is an alternative hypothesis for the low audit specificity [4] [14]. A study reported that among ≤ 3-mm colorectal lesions that were all endoscopically diagnosed as high-confidence adenomas by expert endoscopists, pathology reported up to 15% as normal mucosa [15].

Although interobserver agreement was moderate overall, very high inter-expert agreement was found for high-confidence adenoma diagnoses. Our results underscore the necessity of stating a confidence level when performing optical diagnosis. Good image and video quality for documentation is essential for high-confidence diagnoses to be made. For all experts, high-confidence diagnoses were found to be significantly correlated with good quality polyp presentation (P < 0.001) (Supplementary Table 1). Conversely, poor image and video quality and low confidence level could become a barrier for implementing expert audit. Stability of scope position during polyp visualization and adequate colonoscope set-up and training of endoscopists, therefore, are crucial to enhance picture quality. Moreover, artificial intelligence (AI) could eventually be used to provide real-time feedback to the endoscopist about quality of polyp visualization [16]. Further assessment of expert inter-observer agreement for picture quality documentation is required in future studies assessing CADx.

Overall expert agreement was lower than that reported in previous studies evaluating performance of experienced endoscopists in differentiating neoplastic from hyperplastic lesions, with kappa scores ranging from 0.67 to 0.90 [17] [18]. Inclusion of SSLs, normal mucosa, and inflammatory polyps in the expert review likely plays a role in this decrease in overall agreement; however, when foregoing histopathology within CADx implementation studies this seems to be important. For instance, difficulties frequently arise during endoscopic classification of hyperplastic polyps and SSLs owing to overlapping features between these polyp types, and this could lead to a lack of concordance within studies [1] [19]. Despite recent development of criteria to help endoscopists optically diagnose SSLs, diagnostic accuracy remained low even after training [20]. Poor-to-moderate interobserver agreement between pathologists for differentiating these lesions has also been documented, with one study reporting a kappa value of 0.44 for SSL diagnosis [21]. Moreover, studies have shown moderate-to-high agreement among pathologists for differentiation of neoplastic versus hyperplastic polyps (kappa 0.67–0.89), which is not ideal considering its use as a gold standard diagnosis [22] [23] [24]. Therefore, our study suggests that a similar performance between expert review and pathology can be expected for diminutive polyp diagnosis even when including SSLs. Thresholds for interobserver agreement and diagnostic accuracy, sensitivity, and specificity should be further defined to ensure sufficient quality for audit of both optical diagnosis and CADx-assisted optical diagnosis.

Because several efforts are currently being made to implement optical diagnosis with the “resect and discard” strategy in real-world settings, an efficient and effective audit strategy becomes essential [7]. For instance, implementation has already begun within the Bowel Cancer Screening Programme in England, and cost and time savings have already been made [25]. To our knowledge, this is the first study to evaluate and support the feasibility of expert audit as a diagnostic reference for assessing CADx performance while also providing an optimal strategy for audit implementation. When adenoma diagnoses are concordant and performed with high confidence, audit by two experts and a third in case of disagreement could be used to measure diagnostic performance of optical diagnosis and CADx-assisted optical diagnosis in upcoming true implementation studies. Further research is needed to generalize the results to multiple experts, because only three were included in this study. Factors that influence the level of confidence in optical diagnosis during audit, namely image and video quality, also should be further analyzed. Moreover, optical review by expert endoscopists could avoid inaccurate histological diagnoses that occur frequently in pathology through errors in polyp retrieval, sectioning, and misdiagnoses as normal mucosa [4]. Errors in histopathology assessments could hinder training and testing of CADx programs by causing difficulties in supplying the “truth” to CADx algorithms [13]. It should be noted that evaluation of expert accuracy in this study was done by using histopathology assessment as the gold standard and was not adjusted for potentially inaccurate pathology results, namely among normal mucosa diagnoses. The calculated performance metrics of expert endoscopists could therefore be underestimated.

The primary strength of this study is inclusion of a large number of polyps for evaluation in a prospective audit. However, the study has several limitations. First, only three experts completed the audit, which limits generalizability of the results. Interobserver variability between individual expert accuracies could potentially affect validity of an expert audit of colorectal polyps. However, our study found that Strategy B, involving two experts completing the audit with a third expert available for arbitration if needed, standardized and increased specificity of the audit so that overall performance metrics remained satisfactory. Studies that compare accuracies of a larger number of experienced endoscopists from different countries are required to evaluate expert interobserver variability in polyp optical diagnosis. Second, intra-observer agreement for each expert was not assessed in this study. Diagnostic performance of each expert could potentially vary and affect overall performance of the expert audit if the review was to be later performed with the same set of polyps, warranting further evaluation. Previous studies have shown moderate-to-high intra-observer agreement for experts predicting polyp histology with a kappa score of 0.61–0.87, which was significantly lower than the perfect intra-observer agreement achieved by CADx [18] [26]. Nevertheless, intra-observer agreement of pathologists for polyp histology has not been well documented in the literature [27]. In a study assessing intra-observer variability among experienced pathologists for colorectal adenomas, a kappa score of 0.41–0.90 was reported for determining histological subtype (tubular, tubulovillous, villous) and grade of dysplasia [28]. Third, the audit structure proposed in this study may be difficult to implement in routine clinical practice especially within large-scale audits, because using experts for manual polyp auditing is impractical and resource intensive. A cost-based analysis comparing expert review to traditional pathology should be performed in future studies to assess sustainability of an audit by expert endoscopists. Furthermore, a solution to impracticability of expert audit would be to train an independent AI model using still images of polyps. An AI-driven audit system could be used as the reference standard to assess accuracy of pathology, expert optical diagnosis, and current performance of CADx-assisted optical diagnosis. Similar to an audit by experts, an AI tool would help avoid inaccuracies that occur in pathology diagnoses due to errors in specimen retrieval and sectioning. Future studies should evaluate the feasibility of AI as an auditing tool to support the practical rollout of optical diagnosis with the “resect and discard” strategy.

 

Scroll to top

Conclusions

In summary, the findings from this study showed high sensitivity of expert review for diagnosing high-confidence adenomas that would be sufficient for assessing CADx-assisted optical diagnosis within clinical implementation studies. A review by two independent, blinded experts, with a third expert available for arbitration if needed, should be considered because it provided adequate diagnostic performance while including all polyps for audit, unlike the single-, two- and three-expert review with concordant, high-confidence diagnoses. Despite only moderate concordance between the experts overall, interobserver agreement was very high for high-confidence diagnoses, especially for adenomas.

 

Scroll to top

 

Scroll to top

Conflict of Interest

Daniel von Renteln has received research funding from ERBE Elektromedizin GmbH, Ventage, Pendopharm, Satisfai, Fujifilm, and Pentax, and has received consultant or speaker fees from Boston Scientific Inc., ERBE Elektromedizin GmbH, Medtronic, and Pendopharm. Roupen Djinbachian has received speaker fees from Fujifilm. The remaining authors declare that they have no conflicts of interest.

Acknowledgement

Daniel von Renteln is supported by a “Fonds de Recherche du Québec Santé” career development award. Roupen Djinbachian is supported by a “Fonds de Recherche du Québec Santé/Ministère de la Santé et des Services Sociaux” clinical research award. The authors would like to thank Samira Hanin (Research Coordinator), Ghislaine Ahoua (Research Assistant), Dina Lasfar (Research Assistant), and Thomas-Andrew Sully Guerrier (Research Assistant) for their contributions to this study.

• References

• 1 Djinbachian R, Rex DK, von Renteln D. Optical polyp diagnosis in the era or artificial intelligence. Am J Gastroenterol 2024; 120: 1268-1274
• 2 Kessler WR, Imperiale TF, Klein RW. et al. A quantitative assessment of the risks and cost savings of forgoing histologic examination of diminutive polyps. Endoscopy 2011; 43: 683-691
• 3 Rex DK, Kahi C, O’Brien M. et al. The American Society for Gastrointestinal Endoscopy PIVI (Preservation and Incorporation of Valuable Endoscopic Innovations) on real-time endoscopic assessment of the histology of diminutive colorectal polyps. Gastrointest Endosc 2011; 73: 419-422
• 4 Ahmad A, Moorghen M, Wilson A. et al. Implementation of optical diagnosis with a "resect and discard" strategy in clinical practice: DISCARD3 study. Gastrointest Endosc 2022; 96: 1021-1032.e2
• 5 Barua I, Wieszczy P, Kudo SE. et al. Real-time artificial intelligence-based optical diagnosis of neoplastic polyps during colonoscopy. NEJM Evid 2022; 1: EVIDoa2200003
• 6 Weigt J, Repici A, Antonelli G. et al. Performance of a new integrated computer-assisted system (CADe/CADx) for detection and characterization of colorectal neoplasia. Endoscopy 2022; 54: 180-184
• 7 Taghiakbari M, Rex DK, Pohl H. et al. Pragmatic resect and discard implementation using computer-assisted optical polyp diagnosis. Gastroenterology 2025; 168: 154-156.e2
• 8 Bisschops R, East JE, Hassan C. et al. Advanced imaging for detection and differentiation of colorectal neoplasia: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - update 2019. Endoscopy 2019; 51: 1155-1179
• 9 Cohen JF, Korevaar DA, Altman DG. et al. STARD 2015 guidelines for reporting diagnostic accuracy studies: explanation and elaboration. BMJ Open 2016; 6: e012799
• 10 Djinbachian R, Haumesser C, Taghiakbari M. et al. Autonomous artificial intelligence vs artificial intelligence-assisted human optical diagnosis of colorectal polyps: a randomized controlled trial. Gastroenterology 2024; 167: 392-399.e2
• 11 WHO Classification of Tumours Editorial Board. Digestive System Tumours, WHO Classification of Tumours. Lyon, France: International Agency for Research on Cancer; 2019
• 12 Houwen BBSL, Hassan C, Coupé VMH. et al. Definition of competence standards for optical diagnosis of diminutive colorectal polyps: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy 2022; 54: 88-99
• 13 Rex DK, Bhavsar-Burke I, Buckles D. et al. Artificial intelligence for real-time prediction of the histology of colorectal polyps by general endoscopists. Ann Intern Med 2024; 177: 911-918
• 14 Shahidi N, Rex DK, Kaltenbach T. et al. Use of endoscopic impression, artificial intelligence, and pathologist interpretation to resolve discrepancies between endoscopy and pathology analyses of diminutive colorectal polyps. Gastroenterology 2020; 158: 783-785.e1
• 15 Ponugoti P, Rastogi A, Kaltenbach T. et al. Disagreement between high confidence endoscopic adenoma prediction and histopathological diagnosis in colonic lesions ≤ 3 mm in size. Endoscopy 2019; 51: 221-226
• 16 Ahmad A, Saunders BP. Photodocumentation in colonoscopy: the need to do better?. Front Gastroenterol 2021; 13: 337-341
• 17 Rogart JN, Jain D, Siddiqui UD. et al. Narrow-band imaging without high magnification to differentiate polyps during real-time colonoscopy: improvement with experience. Gastrointest Endosc 2008; 68: 1136-1145
• 18 Chen PJ, Lin MC, Lai MJ. et al. Accurate classification of diminutive colorectal polyps using computer-aided analysis. Gastroenterology 2018; 154: 568-575
• 19 Kim DH, Fournier S, Medawar E. et al. Prospective video-based study assessing effect of computer-assisted optical diagnosis on distinguishing serrated, hyperplastic, and adenomatous colorectal polyps. Digest Dis Sci 2025; 70: 1477-1485
• 20 Soons E, Bisseling TM, van der Post RS. et al. The Workgroup Serrated Polyps and Polyposis (WASP) classification for optical diagnosis of colorectal diminutive polyps with iScan and the impact of the revised World Health Organization (WHO) criteria. United European Gastroenterol J 2021; 9: 819-828
• 21 Vennelaganti S, Cuatrecasas M, Vennalaganti P. et al. Interobserver agreement among pathologists in the differentiation of sessile serrated from hyperplastic polyps. Gastroenterology 2021; 160: 452-454.e1
• 22 Yoon H, Martin A, Benamouzig R. et al. Inter-observer agreement on histological diagnosis of colorectal polyps: the APACC study. [Article in French] Gastroenterol Clin Biol 2002; 26: 220-224
• 23 Foss FA, Milkins S, McGregor AH. Inter-observer variability in the histological assessment of colorectal polyps detected through the NHS Bowel Cancer Screening Programme. Histopathology 2012; 61: 47-52
• 24 van Putten PG, Hol L, van Dekken H. et al. Inter-observer variation in the histological diagnosis of polyps in colorectal cancer screening. Histopathology 2011; 58: 974-981
• 25 Ahmad A, Hearing S, Stebbing J. et al. Implementation of optical diagnosis with a resect and discard strategy for diminutive polyps in the English Bowel Cancer Screening Programme. Endoscopy 2025; 57: 115
• 26 Rastogi A, Pondugula K, Bansal A. et al. Recognition of surface mucosal and vascular patterns of colon polyps by using narrow-band imaging: interobserver and intraobserver agreement and prediction of polyp histology. Gastrointest Endosc 2009; 69: 716-722
• 27 Costantini M, Sciallero S, Giannini A. et al. Interobserver agreement in the histologic diagnosis of colorectal polyps. the experience of the multicenter adenoma colorectal study (SMAC). J Clin Epidemiol 2023; 56: 209-214
• 28 Jensen P, Krogsgaard MR, Christiansen J. et al. Observer variability in the assessment of type and dysplasia of colorectal adenomas, analyzed using kappa statistics. Dis Colon Rectum 1995; 38: 195-198

Publication History

Received: 23 April 2025

Accepted: 03 June 2025

Accepted Manuscript online:
10 June 2025

Article published online:
01 July 2025