Preprint
VIDS: A Verified Imaging Dataset Standard for Medical AI
The founder-authored paper introduces the standard and its published documentation-scoring framework.
Read the Paper →VIDS Stewardship
Stewarding the open standard for medical imaging dataset documentation
Princeton Medical Systems stewards VIDS, the Verified Imaging Dataset Standard. VIDS defines how a medical imaging dataset documents its origin, structure, annotations and versions, and ships with an open validator that checks that documentation is present and structured. Published under CC BY 4.0. Governed in public.
VIDS verifies that documentation is present and structured; it does not certify a dataset. It does not judge clinical quality, model performance, bias, authenticity or fitness for a particular use.
The Documentation Gap
Why This Matters
When the VIDS scoring framework was applied to four widely used public datasets in a founder-authored preprint (arXiv 2604.17525), documentation scores ranged from 20% to 39%. The datasets are public and the dimensions are published, so the finding can be checked rather than taken on our word.
That gap between how datasets are used and how they are documented is the problem VIDS exists to make visible.
The Standard
What VIDS covers
Dataset description, participant and imaging metadata, annotation provenance, quality and split documentation, and version identity, across POC and Full profiles.
Explore VIDSThe validator reports PASS, FAIL, WARN or SKIP per rule. It does not judge whether a dataset is good; it reports whether the documentation is there.
What PMS Does
Standards stewardship, research and education
PMS maintains VIDS, publishes work around the standard, and develops practical capability in medical imaging data documentation.
The Standard
Specification, validator and extensions, maintained by the VIDS Technical Steward and released through public governance.
Explore VIDS →Research, Extensions and Reference Implementations
Peer-reviewable work on dataset documentation and provenance, including LIDC-Hybrid-100 and the VIDS-Fundus 1.0 reference implementation, published with persistent DOI records.
Explore Our Work →Education
Courses, internships and university partnerships that train engineers to work with medical imaging data, documentation standards and reproducible data pipelines.
Education →VIDS Community
Who VIDS is for
VIDS gives different participants a shared framework for documenting and reviewing medical imaging datasets.
Dataset developers
For teams that want their documentation to be checkable by anyone who receives the data.
Imaging-AI teams
For organizations that need to know what a dataset does and does not document before they build on it.
Institutions and reviewers
For organizations that assess datasets before procurement, a study or a submission.
Standards and research communities
For communities that want a common, open vocabulary for imaging dataset documentation rather than another proprietary one.
Published Work
Research and Reference Work
Reference Implementation
LIDC-Hybrid-100
A public VIDS-Full reference implementation with supporting documentation and a persistent DOI record.
View on Zenodo →Extension
VIDS-Fundus 1.0
A fundus-imaging extension with a metadata-only reference implementation and persistent DOI records.
View the Extension →