Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 19 of the 19 references it cites were found in a registry or an index; 2 of the 6 sources its claims cite were checked for support.

How AI is used in FDA-authorized medical devices: a taxonomy across 1,016 authorizations

Open access

Published version of a preprint

Abstract

We reviewed 1016 FDA authorizations of AI/ML-enabled medical devices to develop a taxonomy capturing key variations in clinical and AI-related features. Quantitative image analysis remains the most common application, but its relative proportion has declined recently. Over 100 devices leverage AI for data generation, though none yet involve LLMs. Our taxonomy clarifies current AI usage in medical devices and provides a foundation for tracking developments as applications evolve.

Overview

CoChat Summary

TL;DR

The authors reviewed all 1,016 FDA authorizations of AI/ML-enabled medical devices listed as of December 20, 2024 and built a taxonomy around clinical function, AI function, and data type. They found that image-based quantitative analysis dominates, but its share has declined recently; more than 100 devices use AI for data generation, and no authorized devices in the list used LLMs.

Key findings

  • Across 1,016 FDA authorizations, the authors identified 736 unique devices.
  • By data type, 621 devices (84.4%) use images, 107 (14.5%) use signals, 5 (0.7%) use ‘omics data, and 3 (0.4%) use tabular EHR data.
  • By clinical function, 84.1% are Assessment devices and 15.9% are Intervention devices; by AI function, 630 devices (85.6%) are Analysis, 83 (11.3%) are Generation, and 23 (3.1%) are both.
  • Among Analysis devices, 427 (65% of Analysis devices; 58% of all devices) are quantification/feature localization; among Generation devices, 84 (79.2%) are image enhancement, and 106 devices in total perform Generation.
  • Trends over time: image-based devices peaked at 94% of new devices in 2021 and were 81% in 2024 so far; quantification/feature localization peaked at 81% in 2016 and declined to 51% in 2024.
  • The authors report no evidence of large language models (LLMs) in the studied device list.

Methods

  • The study is a descriptive taxonomy study of the FDA’s published AI/ML-enabled medical device list, accessed from an update on December 20, 2024 and covering authorizations up to September 27, 2024.
  • The team reviewed authorization summaries and product codes, then manually assigned each device four taxonomy factors: data type, clinical function, AI function, and AI function subclass.
  • They grouped multiple authorizations into unique devices using semi-automated matching by product code, company name, device name, and indication for use, followed by manual review; trend analyses were done at the device level using the earliest authorization for each device.

Limitations

  • The paper states that authorization summaries provide only a snapshot, so AI/ML use was sometimes ambiguous; the authors therefore emphasize overall trends rather than precise class counts.
  • It states that the summaries often did not reveal the full form or extent of AI/ML involvement, so the taxonomy captures end use rather than the specific AI method or whether the AI function was fully automated versus clinician-adjusted.
  • It states the study relied on manual curation because automated LLM-based classification was insufficient; it also notes dependence on the FDA’s published list, which is updated irregularly.
  • Evident from the text: because the database is based on the FDA list of authorized devices, it does not include devices used through other regulatory pathways such as investigational device exemption or humanitarian device exemption.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41746-025-01800-1

Cites

19 of 19 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Taxonomy of AI/ML-enabled medical devices. a Illustration of the three core factors: data type, AI function, and clinical function. b Distribution of factor counts across FDA-authorized devices.
Fig. 1 Taxonomy of AI/ML-enabled medical devices. a Illustration of the three core factors: data type, AI function, and clinical function. b Distribution of factor counts across FDA-authorized devices.CC BY-NC-ND · PubMed Central
Fig. 2 Subclasses of AI function in FDA-authorized medical devices. a Illustration of the nine subclasses. b Distribution of AI function subclasses across FDA-authorized devices. Counts indicate the number of devices with the AI function subclass.
Fig. 2 Subclasses of AI function in FDA-authorized medical devices. a Illustration of the nine subclasses. b Distribution of AI function subclasses across FDA-authorized devices. Counts indicate the number of devices with the AI function subclass.CC BY-NC-ND · PubMed Central
Fig. 3 Trends in data type and AI function subclass over time. The percentage of new devices per year (*up to Sept. 27 for 2024) is displayed by a data type and b AI function subclass. The AI function subclasses of predictive, synthetic data generation, and acquisition guidance are grouped as “other” given low counts.
Fig. 3 Trends in data type and AI function subclass over time. The percentage of new devices per year (*up to Sept. 27 for 2024) is displayed by a data type and b AI function subclass. The AI function subclasses of predictive, synthetic data generation, and acquisition guidance are grouped as “other” given low counts.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 0 Not checked · 42 Own results and interpretations

15 more claims

Create a free account to see all 44 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 19
  • Probably not a substantial claim. 7

Create a free account to see these claims.

Citations

Cites

Show all 19 references

Cited by

See all 61

Metadata

Authors
Singh, R., Bapna, M., Diab, A. R., Ruiz, E. S., & Lotter, W.
Affiliations
  • Dana-Farber Cancer Institute (US)
  • Brigham and Women's Hospital (US)
Corresponding author
Lotter, William
Venue
npj Digital Medicine
Published
2025-07-01
Volume
8
Issue
1
Pages
388-388
ISSN
2398-6352 (ISSN-L)
OpenAlex
W4411886794
Cited by
61
References
19
Citations by year
2026: 53 · 2025: 8
FWCI
14.56
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by-nc-nd)
  • medRxiv (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Ellison Foundation
  • Wong Family Award
  • National Institute of Biomedical Imaging and Bioengineering (R21EB035247)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Healthcare Regulation
  • Physical Sciences › Computer Science › Information Systems › Internet of Things and AI
Keywords
AI-enabled medical devices, machine learning, artificial intelligence in healthcare, quantitative image analysis, synthetic data generation, large language models

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: How AI is used in FDA-authorized medical devices: a taxonomy across 1,016 authorizations. https://app.cochat.ai/papers/10.1038/s41746-025-01800-1
The paper
Singh, R., Bapna, M., Diab, A. R., Ruiz, E. S., & Lotter, W. (2025). How AI is used in FDA-authorized medical devices: a taxonomy across 1,016 authorizations. npj Digital Medicine. https://doi.org/10.1038/s41746-025-01800-1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.