Artificial Intelligence–Aided Colonoscopy for Characterizing and Detecting Colorectal Polyps: Required, Nice to Have, or Overhyped?

See “Comparative performance of artificial intelligence optical diagnosis systems for leaving in situ colorectal polyps,” by Hassan C, Sharma P, Mori Y, et al, on page 467.See “Computer-aided detection of polyps does not improve colonoscopist performance in a pragmatic implementation trial,” by Ladabaum U, Shepard J, Weng Y, et al, on page 481. See “Comparative performance of artificial intelligence optical diagnosis systems for leaving in situ colorectal polyps,” by Hassan C, Sharma P, Mori Y, et al, on page 467. See “Computer-aided detection of polyps does not improve colonoscopist performance in a pragmatic implementation trial,” by Ladabaum U, Shepard J, Weng Y, et al, on page 481. In recent years, there has been a litany of publications assessing the potential benefits of emerging artificial intelligence (AI) applications in improving select aspects of colonoscopy quality. This has especially been the case for tools to assist in polyp characterization (computer-aided diagnosis [CADx]) and detection (computer-aided detection [CADe]).1Aziz M, Haghbin H, Sayeh W, et al. Comparison of artificial intelligence with other interventions to improve adenoma detection rate for colonoscopy: a network meta-analysis [published online ahead of print November 28, 2022]. J Clin Gastroenterol https://doi.org/10.1097/MCG.0000000000001813.Google Scholar,2Rondonotti E. Hassan C. Tamanini G. et al.Artificial intelligence-assisted optical diagnosis for the resect-and-discard strategy in clinical practice: the Artificial intelligence BLI Characterization (ABC) study.Endoscopy. 2023; 55: 14-22Crossref PubMed Scopus (8) Google Scholar Several high-quality studies confirmed excellent accuracy in predicting polyp pathology when using CADx for optical diagnosis. The overwhelming evidence from published randomized trials of CADe suggests an improvement in adenoma detection rate (ADR) that may be greater than any other endoscopic intervention.1Aziz M, Haghbin H, Sayeh W, et al. Comparison of artificial intelligence with other interventions to improve adenoma detection rate for colonoscopy: a network meta-analysis [published online ahead of print November 28, 2022]. J Clin Gastroenterol https://doi.org/10.1097/MCG.0000000000001813.Google Scholar In this issue of Gastroenterology, two studies raise questions about the actual real-life benefits of such CADx3Hassan C. Sharma P. Mori Y. et al.Comparative performance of artificial intelligence optical diagnosis systems for leaving in situ colorectal polyps.Gastroenterology. 2023; 164: 467-469Abstract Full Text Full Text PDF Scopus (2) Google Scholar and CADe4Ladabaum U. Shepard J. Weng Y. et al.Computer-aided detection of polyps does not improve colonoscopist performance in a pragmatic implementation trial.Gastroenterology. 2023; 164: 481-483Abstract Full Text Full Text PDF PubMed Scopus (2) Google Scholar technologies. Hassan et al3Hassan C. Sharma P. Mori Y. et al.Comparative performance of artificial intelligence optical diagnosis systems for leaving in situ colorectal polyps.Gastroenterology. 2023; 164: 467-469Abstract Full Text Full Text PDF Scopus (2) Google Scholar directly compared two commercially available CADx systems. Such head-to-head comparisons are much needed, as the platforms have usually been trained and validated with different polyp datasets. Polyps were removed and CADx optical diagnoses were compared with pathology results. The hope is that CADx can one day replace conventional histopathologic analysis for diminutive polyps, reducing costs and allowing the practice of “resect and discard” or “leave non-neoplastic recto-sigmoid polyps in situ.”5Hassan C. Antonelli G. Dumonceau J.M. et al.Post-polypectomy colonoscopy surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - Update 2020.Endoscopy. 2020; 52: 687-700Crossref PubMed Scopus (154) Google Scholar,6Rex D.K. 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: 419-422Abstract Full Text Full Text PDF PubMed Scopus (435) Google Scholar Using innovative study methodology, the same polyp was visualized by the endoscopist on two different monitors simultaneously with respective outputs from each of the two CADx platforms. CAD-unaided (human only) diagnoses were also recorded pre- and post-CADx. The two CADx systems showed a high degree of diagnostic concordance. Both systems exceeded the required benchmark performance when adopting the “diagnose and leave” approach in ruling out adenomas among rectosigmoid diminutive polyps. Importantly, this was also the case for CAD-unaided performance.5Hassan C. Antonelli G. Dumonceau J.M. et al.Post-polypectomy colonoscopy surveillance: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - Update 2020.Endoscopy. 2020; 52: 687-700Crossref PubMed Scopus (154) Google Scholar,6Rex D.K. 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: 419-422Abstract Full Text Full Text PDF PubMed Scopus (435) Google Scholar Next, the role of CADx as part of a “resect and discard” approach to colonic diminutive polyps with recommendations on subsequent colonoscopy surveillance intervals was characterized. The European Society of Gastrointestinal Endoscopy guidelines benchmark was satisfied, but not that of the US guidelines. This discrepancy may reflect that CADx works best when adopting a 10-year surveillance interval for hyperplastic and other low-risk polyps. Here too, and perhaps surprisingly to many, unassisted endoscopic diagnosis outperformed CADx, exceeding both the European and US guidelines’ “resect and discard” performance thresholds. This is in keeping with findings from prior studies.7Hassan C. Balsamo G. Lorenzetti R. et al.Artificial intelligence allows leaving-in-situ colorectal polyps.Clin Gastroenterol Hepatol. 2022; 20: 2505-2513.e4Abstract Full Text Full Text PDF PubMed Scopus (6) Google Scholar,8Mori Y. Kudo S.-E. Misawa M. et al.Real-time use of artificial intelligence in identification of diminutive polyps during colonoscopy: a prospective study.Ann Intern Med. 2018; 169: 357-366Crossref PubMed Scopus (259) Google Scholar The high expertise of participating endoscopists may partially explain a lack of benefit when passing from CADx-unassisted to CADx-assisted endoscopist diagnosis in terms of both technical accuracy and clinical outcomes. Exploratory analysis of a recent cohort study suggested lower AI-assisted optical diagnosis accuracy for nonexperts may, over time, approach that of experts2Rondonotti E. Hassan C. Tamanini G. et al.Artificial intelligence-assisted optical diagnosis for the resect-and-discard strategy in clinical practice: the Artificial intelligence BLI Characterization (ABC) study.Endoscopy. 2023; 55: 14-22Crossref PubMed Scopus (8) Google Scholar; this observation requires further study. Another possible limitation of current CADx systems is that they classify polyps into neoplastic and non-neoplastic diagnoses only. Existing marketed CADx solutions as such cannot adequately replace the added granular information provided by pathology reports for serrated, villous, high-grade dysplastic, or cancerous lesions. Future CADx system development should target additional polyp histologies, with resultant real-life outcomes and costs measured accordingly. Next, Ladabaum et al4Ladabaum U. Shepard J. Weng Y. et al.Computer-aided detection of polyps does not improve colonoscopist performance in a pragmatic implementation trial.Gastroenterology. 2023; 164: 481-483Abstract Full Text Full Text PDF PubMed Scopus (2) Google Scholar presented a pragmatic implementation study in routine practice, measuring the impact of CADe on a comprehensive set of colonoscopy quality metrics. The authors assessed the generalizability of published randomized trials of CADe and, in particular, real-life uptake of such technology. The study design of this pre-post trial was adapted to address potential variations resulting from a period effect (possible secular trend) and separate comparator sites (differing patient and endoscopist populations)—independent of CADe introduction—and after adjustment for age, sex, and indication. Introduction of the CADe application resulted in no significant effect on ADR, adenoma per colonoscopy, sessile serrated lesion, or advanced lesion detection rates, procedure times, or non-neoplastic resection rates. Although larger studies with the same findings might reveal some between-arm differences as statistically significant, such small increments would still be in sharp contrast to results of randomized trials, as was the lack of improvement in ADR for lower-tertile colonoscopists. A possible but improbable source of confounding relates to the high baseline withdrawal times, as CADe should have helped low detectors identify those polyps appearing on the screen that could have otherwise been missed. An additional confounder could be the high baseline ADR of colonoscopists, but this too is less likely when considering that there were no substantial improvements among the lower and middle tertiles of endoscopists, and those in the top tertile worsened numerically after CADe implementation. Other reasons for the lack of benefits may relate to growing overreliance over time on CADe notification of possible polyps. These findings are in keeping with recent publications bringing CADe benefits and their magnitude into question.9Levy I. Bruckmayer L. Klang E. et al.Artificial intelligence-aided colonoscopy does not increase adenoma detection rate in routine clinical practice.Am J Gastroenterol. 2022; 117: 1871-1873Crossref PubMed Scopus (4) Google Scholar,10Ahmad A, Wilson A, Haycock A, et al. Evaluation of a real-time computer-aided polyp detection system during screening colonoscopy: AI-DETECT study [published online ahead of print December 12, 2022]. Endoscopy https://doi.org/10.1055/a-1966-0661Google Scholar Perhaps most importantly, the Ladabaum et al study highlights that randomized controlled trial results must now be complemented with information on dissemination and real-life uptake of CAD colonoscopy technology. Such information may best be obtained through pragmatic trial designs, perhaps also focusing on multiple quality metrics, such as adenoma per colonoscopy and adenoma miss rates, not just ADR. Of course, the global benefits of colonoscopy-based CAD platforms can become fully appreciable only once their impact on the breadth of colonoscopy quality facets are rigorously studied. These include assessment of bowel preparation, mucosal surface area observed, polyp size, and completeness of resection, in addition to polyp detection and characterization. As highlighted in both articles, however, appropriate educational initiatives addressing colonoscopy skills and AI solution implementation will also be critical in this emerging era of computer-assisted diagnostic and therapeutic platforms. Only then will we know whether the gastroenterology community and its patients can truly benefit from a comprehensive suite of AI-driven workflow solutions. Computer-aided Detection of Polyps Does Not Improve Colonoscopist Performance in a Pragmatic Implementation TrialGastroenterologyVol. 164Issue 3PreviewArtificial intelligence (AI), including computer-aided detection (CADe), could revolutionize endoscopy. The adenoma detection rate (ADR) is inversely associated with the risk of postcolonoscopy colorectal cancer.1 The first CADe device approved in the United States (GI Genius; Medtronic, Minneapolis, MN) significantly increased the ADR and adenomas per colonoscopy (APC)2,3 and decreased the adenoma miss rate4 in randomized trials. Full-Text PDF Comparative Performance of Artificial Intelligence Optical Diagnosis Systems for Leaving in Situ Colorectal PolypsGastroenterologyVol. 164Issue 3PreviewScreening colonoscopy is effective in reducing colorectal cancer risk but also represents a substantial financial burden.1,2 Novel strategies based on artificial intelligence (AI; computer-aided diagnosis [CADx]) may enable targeted removal only of polyps deemed to be neoplastic, thus reducing patient burden for unnecessary removal of non-neoplastic polyps and reducing costs for histopathology.3–6 The American Society for Gastrointestinal Endoscopy recommends a threshold for optical diagnosis of at least 90% negative predictive value (NPV) for rectosigmoid neoplastic polyps ≤ 5 mm. Full-Text PDF