Language Models Can Assist Technicians Choosing a Patient-Tailored MRI Scan Protocol

Motivation: Selecting MRI scan protocols is a manual task for technicians, requiring extensive scanner-specific training. Current support methods lack precision and adaptability to scan requirements. Goal(s): Develop a tool using a RAG approach, leveraging a language model to suggest optimal protocols and strategies based on site-specific protocol layouts and requested procedures. Approach: Our method incorporates the local protocol tree and RIS layout to streamline protocol selection. A feasibility study, using simulated scan requests and a vendor's protocol tree, assesses its effectiveness. Results: Our RAG-based tool reliably aids protocol selection with up to 97.8% success rate without additional language model training. Impact: This work streamlines MRI scan protocol selection using a context-aware, RAG-based pipeline. By minimizing manual input and training needs, it demonstrates potential for enhancing workflow efficiency and patient-adapted imaging based on available patient information.