Advertisement
Original article Clinical endoscopy: Editorial| Volume 91, ISSUE 2, P340-341, February 2020

Download started.

Ok

Artificial intelligence in endoscopy: the guardian angel is around the corner

      Abbreviations:

      AI (artificial intelligence), C-EGD (conventional EGD), U-TOE (ultrathin transoral endoscopy)
      Artificial intelligence (AI), which mimics human cognitive functioning, has revolutionized many industries, including medicine.
      • Poole D.L.
      • Mackworth A.K.
      Artificial Intelligence.
      Recent advances in technology, including increased computational power, more efficient hardware, and the development of deep learning algorithms, have led to the emergence of several AI applications in GI endoscopy.
      • Alagappan M.
      • Brown J.R.G.
      • Mori Y.
      • et al.
      Artificial intelligence in gastrointestinal endoscopy: the future is almost here.
      Physicians who perform endoscopy often require several clinical skills, including the simultaneous dexterous operation of endoscopic devices and also the visual identification of disease, so as to drive clinical decision making. This is a complex task, and missed lesions remain a well-recognized issue.
      • Menon S.
      • Trudgill N.
      How commonly is upper gastrointestinal cancer missed at endoscopy? A meta-analysis.
      Several tools that use AI have been developed to aid physicians in performing these tasks with applications to improve mucosal visualization, adenoma detection, and characterization.
      • Mori Y.
      • Kudo S.-E.
      • Berzin T.M.
      • et al.
      Computer-aided diagnosis for colonoscopy.
      • Rees C.J.
      • Koo S.
      Artificial intelligence: upping the game in gastrointestinal endoscopy?.
      • Sehgal V.
      • Rosenfeld A.
      • Graham D.G.
      • et al.
      Machine learning creates a simple endoscopic classification system that improves dysplasia detection in Barrett’s oesophagus amongst non-expert endoscopists.
      In the current issue of Gastrointestinal Endoscopy, Chen et al
      • Chen D.
      • Wu L.
      • Li Y.
      • et al.
      Comparing blind spots of unsedated ultrafine, sedated, and unsedated conventional gastroscopy with and without artificial intelligence: a prospective, single-blind, 3-parallel-group, randomized, single-center trial.
      present data on the use of their novel AI system ENDOANGEL, previously named WISENSE, in improving mucosal visualization, in several modalities of EGD, namely, sedated conventional EGD (C-EGD), unsedated C-EGD, and ultrathin transoral endoscopy (U-TOE). This research group has previously demonstrated the efficacy of their AI technology in a randomized controlled trial of 324 patients, with reduced rates of blind spot in comparison with a control group (5.68% vs 22.46%, P < .001).

      Wu L, Zhang J, Zhou W, et al. Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy. Gut. Epub 2019 Mar 11.

      This AI system is based on deep convolutional neural networks, a method that uses machine learning to perform descriptive and generative tasks, using image and video recognition and with the ability to make predictions. It also uses deep reinforcement learning, machine learning that takes action by observing an environment and using self-feedback to improve results.

      Wu L, Zhang J, Zhou W, et al. Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy. Gut. Epub 2019 Mar 11.

      ENDOANGEL provides real-time prompting of blind spot areas using a virtual stomach model, allowing endoscopists to examine these missed parts, thereby improving mucosal visualization. It also informs the endoscopists of the inspection time for each procedure and provides a grading score of “good,” “excellent,” or “perfect” when 80%, 90%, or 100% of the mucosa is visualized.
      This single-center study enrolled 437 patients who were first randomized to an endoscopic modality: C-EGD (n = 146), unsedated C-EGD (n = 146), and U-TOE (n = 145). The patients in each subgroup were then randomized in a 1:1 ratio to have their procedures performed with or without ENDOANGEL technology. In this study, a complete endoscopic examination required visualization of 26 sites, based on a combination of 2 endoscopic guidelines: the Japanese standard protocol, which requires 22 views of the stomach, and the European Society of Gastrointestinal Endoscopy guidelines, which require at least 10 protocol views of the esophagus, gastroesophageal junction, stomach, duodenal bulb, and descending duodenum.
      • Yao K.
      • Uedo N.
      • Muto M.
      • et al.
      Development of an E-learning system for the endoscopic diagnosis of early gastric cancer: an international multicenter randomized controlled trial.
      ,
      • Bisschops R.
      • Areia M.
      • Coron E.
      • et al.
      Performance measures for upper gastrointestinal endoscopy: a European Society of Gastrointestinal Endoscopy (ESGE) quality improvement initiative.
      The videos from all procedures were independently reviewed by 5 endoscopists, and consensus was reached regarding the 26 predetermined blind spots for each patient.
      The overall blind spot rate was lowest in sedated C-EGD compared with both U-TOE and unsedated C-EGD. The use of the ENDOANGEL system reduced the blind spot rate among all 3 endoscopic modalities: C-EGD (3.42% vs 22.36%, P < .001), U-TOE (21.77% vs 29.92%, P < .001) and unsedated C-EGD (31.23% vs 42.46%, P < .001). It reduced the blind spot rate of C-EGD, U-TOE, and unsedated C-EGD by 84.77%, 27.24%, and 26.45%, respectively. Sedated C-EGD was the best tolerated modality as measured by patient-reported comfort scores.
      This study builds on the application of the AI system ENDOANGEL to improve mucosal visualization in EGD. It demonstrates its effectiveness in both sedated and unsedated procedures along with the use of a conventional and ultrathin oral endoscope. A limitation of this system, however, is that it does not enhance the detection of abnormal mucosa or mucosal lesions, which is an important focus of development to improve its clinical application. Although the average duration of endoscopic procedures is provided, the study group does not comment on the effect of ENDOANGEL on total procedure time. Significant increases to procedure time may limit its clinical utility.
      ENDOANGEL has great potential to improve the diagnostic effectiveness of EGD by improving mucosal visualization. It could be used as a quality indicator to assess the completeness of an EGD. We currently have good quality measures in colonoscopy with cecal intubation rates and withdrawal times; however, these measures are currently lacking in upper GI endoscopy. ENDOANGEL also may play a role in endoscopic training programs and in assessing the competence of endoscopists. The development of AI systems in the field of gastroenterology is rapidly expanding and will revolutionize diagnostic endoscopy in the near future.

      Disclosure

      All authors disclosed no financial relationships relevant to this publication.

      References

        • Poole D.L.
        • Mackworth A.K.
        Artificial Intelligence.
        Cambridge University Press, Cambridge2010
        • Alagappan M.
        • Brown J.R.G.
        • Mori Y.
        • et al.
        Artificial intelligence in gastrointestinal endoscopy: the future is almost here.
        World J Gastrointest Endosc. 2018; 10: 239-249
        • Menon S.
        • Trudgill N.
        How commonly is upper gastrointestinal cancer missed at endoscopy? A meta-analysis.
        Endosc Int Open. 2014; 02: E46-E50
        • Mori Y.
        • Kudo S.-E.
        • Berzin T.M.
        • et al.
        Computer-aided diagnosis for colonoscopy.
        Endoscopy. 2017; 49: 813-819
        • Rees C.J.
        • Koo S.
        Artificial intelligence: upping the game in gastrointestinal endoscopy?.
        Nat Rev Gastroenterol Hepatol. 2019; 16: 584-585
        • Sehgal V.
        • Rosenfeld A.
        • Graham D.G.
        • et al.
        Machine learning creates a simple endoscopic classification system that improves dysplasia detection in Barrett’s oesophagus amongst non-expert endoscopists.
        Gastroenterol Res Pract. 2018; 2018: 1-9
        • Chen D.
        • Wu L.
        • Li Y.
        • et al.
        Comparing blind spots of unsedated ultrafine, sedated, and unsedated conventional gastroscopy with and without artificial intelligence: a prospective, single-blind, 3-parallel-group, randomized, single-center trial.
        Gastrointest Endosc. 2020; 91: 332-339.e1-3
      1. Wu L, Zhang J, Zhou W, et al. Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy. Gut. Epub 2019 Mar 11.

        • Yao K.
        • Uedo N.
        • Muto M.
        • et al.
        Development of an E-learning system for the endoscopic diagnosis of early gastric cancer: an international multicenter randomized controlled trial.
        EBioMedicine. 2016; 9: 140-147
        • Bisschops R.
        • Areia M.
        • Coron E.
        • et al.
        Performance measures for upper gastrointestinal endoscopy: a European Society of Gastrointestinal Endoscopy (ESGE) quality improvement initiative.
        Endoscopy. 2016; 48: 843-864

      Linked Article