Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Comment
. 2021 Sep 7;193(35):E1391-E1394.
doi: 10.1503/cmaj.202066. Epub 2021 Aug 30.

Problems in the deployment of machine-learned models in health care

Joseph Paul Cohen  1 , Tianshi Cao  2 , Joseph D Viviano  2 , Chin-Wei Huang  2 , Michael Fralick  2 , Marzyeh Ghassemi  2 , Muhammad Mamdani  2 , Russell Greiner  2 , Yoshua Bengio  2
Affiliations
Comment

Problems in the deployment of machine-learned models in health care

Joseph Paul Cohen et al. CMAJ. .
No abstract available

PubMed Disclaimer

Conflict of interest statement

Competing interests: Joseph Paul Cohen reports receiving CIFAR Artificial Intelligence and COVID-19 Catalyst grants, and a grant from Carestream Health (Stanford University), outside the submitted work. Dr. Cohen also reports that he is Director of the Institute for Reproducible Research, a non-profit organization in the United States. Michael Fralick reports receiving consulting fees from Proof Diagnostics (previously Pine Trees Health), a start-up company developing a CRISPR-based diagnostic test for SARS-CoV-2 infection. No other competing interests were declared.

Figures

Figure 1:
Figure 1:
This figure shows 3 categories of out-of-distribution data, all in the context of training a machine-learned algorithm to read adult chest radiographs (see image C iii). A) Images that are unrelated to the task. B) Images that are incorrectly acquired. C) Images that are not encountered owing to a selection bias in the training distribution (e.g., images with lung cancer lesions and pacemakers were not included in the training set and therefore were unseen during training). C) (iii) Training data that are subject to a selection bias.
See this image and copyright information in PMC

Comment in

  • Evaluation of machine learning solutions in medicine.
    Antoniou T, Mamdani M. Antoniou T, et al. CMAJ. 2021 Sep 13;193(36):E1425-E1429. doi: 10.1503/cmaj.210036. Epub 2021 Aug 30. CMAJ. 2021. PMID: 34462315 Free PMC article. No abstract available.
  • Implementing machine learning in medicine.
    Verma AA, Murray J, Greiner R, Cohen JP, Shojania KG, Ghassemi M, Straus SE, Pou-Prom C, Mamdani M. Verma AA, et al. CMAJ. 2021 Aug 30;193(34):E1351-E1357. doi: 10.1503/cmaj.202434. Epub 2021 Aug 29. CMAJ. 2021. PMID: 35213323 Free PMC article. No abstract available.
  • Biased data lead to biased algorithms.
    Richardson A. Richardson A. CMAJ. 2022 Mar 7;194(9):E341. doi: 10.1503/cmaj.80860. CMAJ. 2022. PMID: 35256392 Free PMC article. No abstract available.

Comment on

References

    1. Verma AA, Murray J, Greiner R, et al. Implementing machine learning in medicine. CMAJ 2021;193:E1351–7. - PMC - PubMed
    1. Liu Y, et al. How to read articles that use machine learning: users’ guides to the medical literature. JAMA 2019;322:1806–16. - PubMed
    1. England JR, Cheng PM. Artificial intelligence for medical image analysis: a guide for authors and reviewers. AJR Am J Roentgenol 2019;212:513–9. - PubMed
    1. Tschandl P, Rinner C, Apalla Z, et al. Human–computer collaboration for skin cancer recognition. Nat Med 2020;26:1229–34. - PubMed
    1. Artificial intelligence and machine learning in software as a medical device. Silver Spring (MD): US Food and Drug Administration; 2021. Available: https://www.fda.gov/medical-devices/software-medical-device-samd/artific... (accessed 2020 May 14).