Use of Large Language Models to enhance Failure Mode and Effects Analysis: A Case Study

Key Points

• The aim is to determine if large language models can augment expert-driven Failure Mode and Effects Analysis in radiation oncology.
• Evaluated four large language models (ChatGPT-4, Gemini) for their ability to identify failure modes.
• Multidisciplinary team included medical physicists, quality assurance engineers, and software developers.
• Used qualitative analysis to assess effectiveness in the Radiation Planning Assistant workflow.
• Large language models identified several novel failure modes not previously recognized by experts.
• LLMs significantly decreased time spent on the analysis process.
• Feedback indicated improved risk identification compared to traditional methods.

Abstract