Elsevier

Gastrointestinal Endoscopy

Volume 92, Issue 4, October 2020, Pages 813-820.e4
Gastrointestinal Endoscopy

Review article
A primer on artificial intelligence and its application to endoscopy

https://doi.org/10.1016/j.gie.2020.04.074Get rights and content

Artificial intelligence (AI) has emerged as a powerful and exciting new technology poised to impact many aspects of health care. In endoscopy, AI is now being used to detect and characterize benign and malignant GI lesions and assess malignant lesion depth of invasion. It will undoubtedly also find use in capsule endoscopy and inflammatory bowel disease. Herein, we provide the general endoscopist with a brief overview of AI and its emerging uses in our field. We also touch on the challenges of incorporating AI into clinical practice, such as workflow integration, data storage, and data privacy.

Section snippets

Artificial Intelligence

What is artificial intelligence (AI)? Unfortunately, most people, not just physicians, have a limited understanding of the term and associated research. In mainstream media, AI has become a term that refers to software mediated automation and complex algorithms meant to replace human cognitive processes. In reality, AI is an umbrella term that can refer to many subfields.1 The emerging applications of AI in health care and beyond are those that use complex algorithms to augment and support the

Key Terms and Concepts

As mentioned, AI is an umbrella term (Table 1). Algorithms based on AI are usually termed “AI systems,” “AI algorithms,” “AI clinical decision support systems,” or “AI platforms.” The ability of AI to learn and perform tasks without user input is termed “machine learning.”6 Data are fed to the algorithm to train it to make predictions on separate data sets. Machine learning, like traditional regression, generates a prediction based on mathematical computations from the provided data. However,

Data, Learning, Validation, and Outcomes

An understanding of the types of data used and their inherent limitations is useful when critiquing AI literature. Data are usually separated into a training set, validation set, and test set. Training sets are fed to the algorithm and used for initial development. Validation sets confirm an acceptable margin of error using certain user-defined algorithm parameters.9 A test set is then used to fully evaluate the algorithm.

Training data sets can be prone to both “selection bias” and

Regulation and Privacy

The U.S. Food and Drug Administration has deemed that AI tools for clinical support will be regulated as medical devices.15 This is reasonable, given the ramifications if these tools were implemented without proper assessment. AI systems are certainly not infallible, and numerous issues regarding clinical safety will need to be addressed before large-scale implementation.16 Many clinically relevant errors will likely arise from selection bias and overfitting, which simply reinforces the need

Potential Downsides of AI Implementation

In addition to regulation and privacy issues, endoscopists should be aware of other potential downsides of AI as it becomes widely implemented. First, there is concern that overreliance on AI tools will result in deskilling of endoscopists. It is important to remember AI tools are not without diagnostic error. Therefore, endoscopists should be clinically vigilant when using such tools and be critical of diagnoses they believe the AI tool may have made in error. This will be particularly

Applications in Endoscopy

In health care, AI has begun to have impact by rapidly and accurately interpreting images, improving workflow, and reducing medical errors.20 Image-based areas, such as pathology, radiology, and endoscopy, are expected to be the first in health care impacted by AI.21 There is significant growing interest in AI’s ability to increase the performance characteristics of endoscopic imaging modalities and allow nonexperts to reach certain diagnostic thresholds.22

The 2 areas of endoscopy in which AI

The Future

The future for AI in our world of GI endoscopy looks very promising indeed. Within the next 5 years, we believe AI systems for endoscopic diagnostic support will become widely clinically available, certainly for colon polyp characterization and detection and very likely for other disease states and anatomies such as early GI cancer and IBD. AI tools will be used in standard optical colonoscopy as well as CE. Knowledge of these tools will also become a focus of educational activities, because it

Conclusion

AI is a truly exciting technology, which will no doubt have a profound impact on all areas of health care. Gastroenterology and endoscopy have already begun to feel the change. We expect that endoscopy will be one of the first areas in all of medicine in which AI is used on a wide-scale basis, given its inherent reliance on imaging. GI endoscopists should be proud that our field is helping set the tone of AI development in medicine.

We hope we have enlightened our readers with a general

References (66)

  • D.G. Vinsard et al.

    Quality assurance of computer-aided detection and diagnosis in colonoscopy

    Gastrointest Endosc

    (2019)
  • Y. Mori et al.

    Simultaneous detection and characterization of diminutive polyps with the use of artificial intelligence during colonoscopy

    VideoGIE

    (2019)
  • Y. Maeda et al.

    Fully automated diagnostic system with artificial intelligence using endocytoscopy to identify the presence of histologic inflammation associated with ulcerative colitis (with video)

    Gastrointest Endosc

    (2019)
  • R. Leenhardt et al.

    A neural network algorithm for detection of GI angiectasia during small-bowel capsule endoscopy

    Gastrointest Endosc

    (2019)
  • M.F. Byrne et al.

    Artificial intelligence and capsule endoscopy: Is the truly “smart” capsule nearly here?

    Gastrointest Endosc

    (2019)
  • A.P. Engelbrecht

    Computational intelligence: an introduction

    (2007)
  • H. Neumann et al.

    Artificial intelligence and the future of endoscopy

    Dig Endosc

    (2019)
  • A.M. Turing

    The essential turing

    (2004)
  • Y. LeCun et al.

    Deep learning

    Nature

    (2015)
  • T. Mitchell et al.

    Machine learning

    Annu Rev Computer Sci

    (1990)
  • D. Michie et al.

    Machine learning

    Neural Stat Class

    (1994)
  • R. Sathya et al.

    Comparison of supervised and unsupervised learning algorithms for pattern classification

    Int J Adv Res Artific Intell

    (2013)
  • I. Goodfellow et al.

    Deep learning

    (2016)
  • J. Lever et al.

    Points of significance: model selection and overfitting

    (2016)
  • M.F. Byrne et al.

    Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model

    Gut

    (2019)
  • Clinical Decision software: guidance for industry and Food and Drug Administration staff

    (2019)
  • R. Challen et al.

    Artificial intelligence, bias and clinical safety

    BMJ Qual Saf

    (2019)
  • K. Abouelmehdi et al.

    Big healthcare data: preserving security and privacy

    J Big Data

    (2018)
  • W. Burke et al.

    Informed consent in translational genomics: insufficient without trustworthy governance

    J Law Med Ethics

    (2018)
  • E.J. Topol

    High-performance medicine: the convergence of human and artificial intelligence

    Nat Med

    (2019)
  • Cited by (0)

    DISCLOSURE: The following authors disclosed financial relationships: M. F. Byrne: Chief executive officer and shareholder in Satisfai Health; founder of AI4GI joint venture; co-development agreement between Olympus America and AI4GI in artificial intelligence and colorectal polyps. All other authors disclosed no financial relationships.

    View full text