Supporting MRI Technicians: An LLM-Based Troubleshoot Companion for Operational Assistance

Motivation: Operating MRI scanners is complex, and technicians frequently encounter technical issues that disrupt the workflow. Existing keyword-based support tools often fail to provide adequate assistance. Goal(s): Develop an advanced troubleshoot companion (TSC) leveraging large language models (LLMs) to provide faster, context-aware solutions for MRI technicians. Approach: We implemented a retrieval-augmented generation (RAG) system, combining GPT-4 with a curated knowledge base, and evaluated it against existing tools in a user study. Results: The RAG-based TSC was rated more effective and time-efficient, demonstrating potential as a superior support tool for MRI troubleshooting. Impact: This work enhances MRI troubleshooting by introducing a context-aware support tool based on LLMs, improving problem-solving efficiency for technicians. It highlights the potential of RAG systems in healthcare to replace traditional keyword-based search methods with more intelligent solutions.