RSNA ATLAS Documentation
RSNA ATLAS (the Annotated Library of AI Systems) is a hub for the imaging AI research community to share critical information. Researchers and developers can publish data cards that describe the key attributes of their AI models and datasets. Users can use ATLAS to discover, evaluate and compare these resources.
ATLAS Card Creator
Use the ATLAS card creator to develop a model or dataset card for submission to ATLAS
Access the ATLAS card creator to create a card for publication on ATLAS
Expert Panel
The ATLAS data schema and ROADMAP ontology are maintained by a panel of imaging AI experts
ROADMAP (the Radiology Ontology of AI Datasets, Models and Projects) provides a controlled terminology for the metadata describing AI models and datasets. Explore the ROADMAP ontology on the National Center for Biomedical Ontology's BioPortal site.
How It Works
1. Users can submit a "card" that describes the attributes of AI models and/or datasets through our standardized web interface.
2. The repository validates submitted documents against the predefined JSON Schema and checks that all referenced URLs are "live".
3. Expert radiologist reviewers validate model dataset cards to ensure quality and accuracy.
4. Each published "model card" is issued a DOI (Digital Object Identifier) for permanent identification and citation.
5. Characteristics are indexed such as: Project name, model name, dataset name, RadLex code and RSNA content code for enhanced discoverability.
6. Users can search and retrieve JSON-language index cards and access model cards via web interface and API.
The Problem
It is essential to have a searchable catalog that enables users to locate related models and datasets while ensuring their information is interoperable. Identifying pertinent AI models and datasets can be a complex task. There is currently no standard way to express the source data attributes and other relevant information about imaging AI models and no central place for users to research them.
Our Solution
RSNA has developed a hub of standardized "model cards" for AI models and training datasets, establishing a centralized platform for publishing these card documents. The RSNA AI Imaging ROADMAP includes a reference ontology for relevant terminology, tools to produce standardized model cards, and a centralized platform for AI model developers and users.
Key Benefits
For Members, Researchers and Developers
- • Expedited development and integration of AI resources
- • Standardized way to document and share AI models
- • Increased visibility and discoverability of work
For Regulators and Implementers
- • Regulatory Science Tool supporting FDA requirements
- • Transparent information for evaluation and approval
Contact
If you have any questions or comments, please contact: