From Bench to Bedside With Large Language Models: AJR Expert Panel Narrative Review

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Informatics

Background:

• Large language models (LLMs) present transformative potential for the field of radiology.
• Current general-purpose LLMs and AI chatbots face challenges in privacy, transparency, and accuracy, hindering clinical adoption.
• The development and validation of LLM-specific tools for radiology are still in early stages, despite emerging commercial products.

Purpose of the Study:

• To provide a comprehensive, multidimensional review of LLMs in radiology, from development to clinical application.
• To guide radiologists in navigating the evolving landscape of LLM integration.
• To highlight considerations for responsible implementation to maximize benefits and mitigate risks.

Main Methods:

• Narrative review methodology.
• Expert panel insights.
• Multidimensional perspective encompassing bench and bedside considerations.

Main Results:

• LLMs are not currently capable of autonomous decision-making; radiologists retain responsibility for report content.
• Patient-facing AI tools, such as medical chatbots, necessitate robust safety measures and oversight.
• Responsible implementation of LLMs can significantly improve efficiency and quality in radiological practice.

Conclusions:

• LLMs hold promise for enhancing radiology efficiency and quality if implemented responsibly.
• Radiologists must be actively involved in guiding the integration of LLMs to ensure patient safety and optimal outcomes.
• Further research and validation are crucial for LLM applications in clinical radiology.