Tools and Guardrails for Sharing AI Resources at Scale
The landscape of radiology is evolving at a rapid pace, driven largely by advancements in artificial intelligence (AI). The Radiological Society of North America (RSNA) has introduced ATLAS, a groundbreaking platform designed to facilitate the accessibility and interoperability of AI resources.
According to Charles E. Kahn Jr., MD, MS, a key member of the RSNA AI Committee, ATLAS serves as a vital bridge for radiologists, researchers, and developers, making it easier to locate the AI models and datasets that fit their specific needs. This initiative aims to simplify the process of sharing AI resources, thereby enhancing collaboration and innovation in the field of medical imaging.
Features and Tools of ATLAS
One of the standout features of ATLAS is the ATLAS Card Creator, which simplifies the process of creating detailed "index cards" for various AI models and datasets. This user-friendly tool allows individuals to populate a card with essential information seamlessly. An AI extractor tool is also included, designed to pull pertinent data from existing documents, substantially reducing the time and effort needed for card submission.
Moreover, the ATLAS platform maintains strict validation protocols for submissions, ensuring that only high-quality resources are made available to the public. Here are some key aspects of the validation process:
- Model Cards Validation: Each model card is checked against a JSON schema and verified to ensure it contains active URLs.
- Searchable Interface and API: Users can explore model cards through an intuitive web interface or utilize an API to integrate these resources into their own systems effortlessly.
- Ontology-Driven Indexing: Model cards are tagged using RadLex and RSNA content codes, enabling highly accurate categorization.
- Digital Object Identifier (DOI) Assignment: Each published card receives a DOI, promoting traceability and ease of citation.
ROADMAP: An Essential Component
Another important element that bolsters the effectiveness of ATLAS is the Radiology Ontology of AI Datasets, Models, and Projects (ROADMAP). This controlled vocabulary system standardizes the metadata associated with AI models and datasets. The rigorous maintenance of the ATLAS data schema and the ROADMAP ontology is managed by a panel of imaging AI experts, assuring that the information shared is both reliable and up to date.
Dr. Kahn emphasizes, “ATLAS utilizes widely accepted vocabularies, such as SNOMED and RadLex, to index its content effectively.” The incorporation of language from the ROADMAP provides additional depth and clarity, helping to standardize the information surrounding available AI resources.
A Call to the Imaging AI Community
The ATLAS platform welcomes contributions from the imaging AI community, encouraging professionals to publish their own ATLAS cards for AI models and datasets they wish to share. By submitting these cards, contributors ensure their work is discoverable and accessible to the global medical imaging and radiology community. This, in turn, facilitates easier understanding, evaluation, and potential collaboration for real-world applications.
Explore Further
For anyone interested in diving deeper into the capabilities of ATLAS or wanting to explore additional resources, RSNA has made it easy to access a wealth of information:
- Learn more about ATLAS now.
- Explore additional RSNA AI resources, which include datasets and tools designed for model development.
- Discover peer-reviewed research on imaging AI in Radiology: Artificial Intelligence.
- See how RSNA promotes innovation through AI Challenges.
- Read previous stories on data standards and interoperability from RSNA News.
Through initiatives like ATLAS, RSNA continues to pave the way for the effective integration of AI in radiology, ensuring professionals have the tools they need to enhance patient care and streamline workflows.
