Artificial intelligence complemented by water exchange for right-sided colonic polyp detection: It’s time to dive!

Abstract

Artificial intelligence (AI) was introduced to the endoscopy community just less than a decade ago, and it has rapidly become one of the most attractive and promising tools in modern endoscopy. In particular, computer-aided detection (CADe) of colorectal polyps has already been assessed in multiple studies around the world, showing significant improvements in adenoma and polyp detection rates.1Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (118) Google Scholar Another reason behind the success of AI in endoscopy is that it is very easy to use. CADe software can be either integrated with the endoscopy system or implemented via an external platform that can be device agnostic, allowing application across multiple imaging modalities and processors. This software highlights the presence of a suspected polyp by the use of a visual (a box or tag of various colors) and/or an audible signal. The endoscopist in this case is required only to perform a closer inspection of the suspected lesion and verify its nature. If this were not enough, computer-aided characterization (CADx) for polyp characterization can also be used, when available in the endoscopy unit, to support optical diagnosis and facilitate therapeutic decision making. It is intuitive that CADe makes our life easier, with consequent advantages for our patients. We have now another strong ally in the fight against colorectal cancer, which unfortunately is still the fourth most commonly diagnosed and the third most fatal cancer worldwide. Of course, the role of the endoscopist is still very central, and we cannot completely rely on CADe. A careful inspection of the colonic mucosa, especially in traditional blind spots (such as the hepatic and splenic flexures, or behind colonic folds) is still essential, in combination with a controlled and slow colonoscope withdrawal technique alongside optimal bowel preparation. These measures are essential in reducing interval colorectal cancer rates. Another technique that has been proved to be effective in this regard is the water exchange (WE) insertion method: a continuous WE consisting of infusion and suction of the water in an airless bowel during the colonoscopy insertion phase. This has the advantage of reducing patient discomfort, improving bowel preparation, and increasing adenoma detection rates (ADRs).2Cadoni S. Ishaq S. Hassan C. et al.Water-assisted colonoscopy: an international modified Delphi review on definitions and practice recommendations.Gastrointest Endosc. 2021; 93: 1411-1420Abstract Full Text Full Text PDF PubMed Scopus (20) Google Scholar In this issue of Gastrointestinal Endoscopy, Tang et al3Tang C.-P. Lin T.-L. Hsieh Y.-H. et al.Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflation.Gastrointest Endosc. 2022; 95: 1198-1206.e6Abstract Full Text Full Text PDF Scopus (3) Google Scholar report the development of CADe and CADx software and evaluate its performance on retrospective videos from a randomized controlled trial, comparing WE and air insufflation withdrawals in the right side of the colon. The additional polyp detection rate (APDR) of right-sided colonic polyps and number of false positive results per colonoscopy were used as outcomes. The authors developed 2 separate algorithms, respectively, for computer-aided detection (CADe) and characterization (CADx) for colonic polyps. After training and testing, they evaluated the algorithm on 245 videos after successful cecal intubation, previously recorded from a randomized controlled trial (RCT) that evaluated right-sided colon withdrawals with the use of air insufflation and WE.4Tseng C.-W. Hsieh Y.-H. Koo M. et al.Comparing right colon adenoma detection rate during water exchange and air insufflation: a double-blind randomized controlled trial.Tech Coloproctol. 2022; 26: 35-44Crossref Scopus (4) Google Scholar Two endoscopists were involved in the recorded procedures; endoscopist A initially performed a colonoscopy with either WE or air insufflation. Once the cecum was reached, the blinded endoscopist, B, took over the colonoscope and performed a right-sided colon inspection with air insufflation, followed by a second inspection completed by endoscopist A with WE. Two senior expert endoscopists reviewed the videos with CADe overlays to establish a consensus on whether the bounding boxes correctly identified polyps. If this was judged to be an additional polyp identified by CADe, then CADx was applied to determine whether this was adenoma. False positive results (FPs) were also classified by reviewers’ consensus. Tang et al3Tang C.-P. Lin T.-L. Hsieh Y.-H. et al.Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflation.Gastrointest Endosc. 2022; 95: 1198-1206.e6Abstract Full Text Full Text PDF Scopus (3) Google Scholar found that both polyp detection rates and additional polyp detection rates were significantly higher in the WE group than in the air insufflation group (53.7% vs 38.5%, P = .021) and (30.1% vs 12.35, P = .001), respectively, when CADe was applied. Moreover, ADR (41.5% vs 24.6%, P = .006) and APDR (17.1% vs 5.7%, P = .008) were also significantly higher in the WE group. In addition, this study showed that the mean number of false positive results per colonoscopy due to feces (2.13 [1.85] vs 2.90 [2.36], P = .007) and bubbles (0.53 [0.89] vs 1.25 [2.45], P = .001) were significantly lower in the WE group than in the air insufflation group. The major contribution of this study by Tang et al3Tang C.-P. Lin T.-L. Hsieh Y.-H. et al.Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflation.Gastrointest Endosc. 2022; 95: 1198-1206.e6Abstract Full Text Full Text PDF Scopus (3) Google Scholar is the demonstration that adjunctive techniques can further augment CADe performance. Most studies to date have evaluated CADe in isolation.1Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (118) Google Scholar Perhaps one of the major limitations of current iterations of regulatory agency-approved CADe software is the sole focus on polyp recognition while not addressing mucosal exposure technique. Therefore, CADe performance to some degree may also be largely dependent on the endoscopist’s technique. In this study, the use of WE not only increased APDR but also reduced FPs resulting from feces and air bubbles. This synergy between CADe and WE helps to address both major mechanisms contributing to missed polyps while also potentially improving user experience. As AI research and development in endoscopy continues to grow exponentially, careful study design and evaluation is going to be critical for widespread clinical implementation. The study by Tang et al3Tang C.-P. Lin T.-L. Hsieh Y.-H. et al.Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflation.Gastrointest Endosc. 2022; 95: 1198-1206.e6Abstract Full Text Full Text PDF Scopus (3) Google Scholar was a retrospective in-silico study, evaluating the algorithm on previously captured videos. Such studies have been indispensable, especially in the early translational stages for AI in endoscopy. However, important limitations exist, such as the lack of criterion-standard histopathologic analysis for apparent CADe polyp detection, instead relying on endoscopists’ consensus based on visual appearance. Furthermore, AI-endoscopist interaction is not assessed. Ultimately, the benefit of CADe algorithms will rely on endoscopists responding appropriately to CADe outputs while not developing apparent user fatigue due to FPs. This will require evaluation in prospective clinical studies. To date, RCTs for CADe have been promising, demonstrating overall increases in ADR.1Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (118) Google Scholar In addition, FPs have not led to significant increases in withdrawal time. However, the real-world impact of FPs needs to be evaluated further, particularly in nonexpert settings. Standardized classification of FPs, focusing on clinical relevance, would be beneficial for evaluation in trials and also for benchmarking algorithms. The recent publication and validation of such a scheme is an important development in the field.5Spadaccini M, Hassan C, Alfarone L, et al. Comparing Number and relevance Of false activations between two Artificial Intelligence CADe SystEms: the NOISE study. Gastrointest Endosc. Epub 2022 Jan 4.Google Scholar Moreover, the current study reports that the APDR associated with CADe use was limited largely to diminutive adenomas. To demonstrate greater impact, future CADe research should focus on the additional detection of advanced subtle lesions that may easily be overlooked and may contribute to interval colorectal cancer, which has been highlighted in preclinical studies.6Ahmad OF, González-Bueno Puyal J, Brandao P, et al. Performance of artificial intelligence for detection of subtle and advanced colorectal neoplasia. Dig Endosc. Epub 2021 Nov 8.Google Scholar The study by Tang et al3Tang C.-P. Lin T.-L. Hsieh Y.-H. et al.Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflation.Gastrointest Endosc. 2022; 95: 1198-1206.e6Abstract Full Text Full Text PDF Scopus (3) Google Scholar provides a glimpse into an exciting future, where both established and novel techniques and innovations in endoscopy will combine harmoniously alongside AI algorithms in the ongoing quest to drive up quality in endoscopy. A more remarkable breakthrough will likely occur with more advanced algorithms providing real-time feedback on withdrawal technique, ultimately outputting objective, computer vision–based mucosal inspection scores. In fact, early iterations of such quality improvement software deployed alongside CADe algorithms have already demonstrated benefit in RCTs.1Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (118) Google Scholar The endoscopy community now needs to prepare for the rapid pace of AI innovation, with a growing number of algorithms set to be deployed in clinical practice in the near future. There is an urgent need for dedicated AI working groups, ideally led by professional societies, to provide guidance on standardized trial designs, optimal endpoints, and benchmarks for AI use in cases in endoscopy.7Ahmad O.F. Mori Y. Misawa M. et al.Establishing key research questions for the implementation of artificial intelligence in colonoscopy: a modified Delphi method.Endoscopy. 2021; 53: 893-901Crossref PubMed Scopus (20) Google Scholar The future is here; after years of hype, GI endoscopy now has some of the most translationally mature and rigorously evaluated AI applications in healthcare. This leaves us to now ponder how to harness the maximum potential of algorithms in combination with the rest of the expanding range of devices and techniques available for endoscopy. Both authors disclosed no financial relationships. Polyp detection and false-positive rates by computer-aided analysis of withdrawal-phase videos of colonoscopy of the right-sided colon segment in a randomized controlled trial comparing water exchange and air insufflationGastrointestinal EndoscopyVol. 95Issue 6PreviewWater exchange (WE) improves lesion detection but misses polyps because of human limitations. Computer-aided detection (CADe) identifies additional polyps overlooked by the colonoscopist. Additional polyp detection rate (APDR) is the proportion of patients with at least 1 additional polyp detected by CADe. The number of false positives (because of feces and air bubble) per colonoscopy (FPPC) is a major CADe limitation, which might be reduced by salvage cleaning with WE. We compared the APDR and FPPC by CADe between videos of WE and air insufflation in the right-sided colon. Full-Text PDF