Of Rags & Regs : A Practical Retrieval Augmented Generation Artificial Intelligence Agent to Support Ionizing Radiation Use Compliance with Regulations and Accreditation
Abstract
Purpose
A Retrieval-Augmented Generation (RAG) Artificial Intelligence (AI) Agent was developed to accurately, reliably, and quickly address queries and prompts that pertain to regulatory and accreditation compliance, specific to the use of ionizing radiation under the purview of the imaging medical physics and radiation safety program at the University of Miami Health System and the Jackson Health System.
Methods
Publicly available documents relevant to the use of ionizing radiation, in combination with institutional databases, were used to develop a RAG AI Agent powered by the ChatGPT Large Language Model (LLM). Regulatory and accreditation documents included the Florida Administrative Code, US Nuclear Regulatory Commission, US Food and Drug Administration, US Department of Labor, and the American College of Radiology Mammography Accreditation Program. Institutional databases included fully de-identified occupational radiation dose records, radiation protective apparel inventories, imaging equipment performance evaluations, and mammography quality control logs.
Results
A practical non-interpretative RAG AI Agent was developed to address LLM queries and prompts pertaining to regulatory and accreditation compliance, specifically relevant to the use of ionizing radiation. Results of sample prompts and queries were obtained and will be presented. Extemporaneous prompts from the audience will also be solicited to demonstrate the utility of the RAG AI Agent.
Conclusion
A RAG AI Agent was developed to accurately, reliably, and quickly address institution-specific queries and prompts pertaining to regulatory and accreditation compliance, specifically relevant to the use of ionizing radiation.