Large Language Models: A Guide for Radiologists.

Area of Science:

• Artificial Intelligence in Medicine
• Natural Language Processing Applications
• Radiology Informatics

Background:

• Large language models (LLMs) demonstrate significant capabilities beyond traditional natural language processing.
• Pre-training on vast datasets enables LLMs to perform general and domain-specific tasks, including in radiology, without fine-tuning.

Purpose of the Study:

• To provide radiologists with conceptual knowledge and practical guidance on utilizing LLMs.
• To offer a concise overview of LLMs and their specific relevance to the field of radiology.
• To summarize current applications and explore potential future directions for LLMs in radiology.

Main Methods:

• Review of current large language model capabilities and their application in medical fields.
• Analysis of how general-purpose LLM-based chatbots can enhance radiologist workflow and research.
• Discussion of ongoing LLM evolution, including solutions for hallucination, cost, efficiency, and multimodal inputs.

Main Results:

• LLMs can optimize radiologist efficiency in professional tasks and research endeavors.
• Contemporary LLMs possess the capacity for domain-specific applications, such as in radiology, without requiring additional fine-tuning.
• Rapid LLM evolution is addressing key challenges and incorporating multimodal capabilities.

Conclusions:

• LLMs represent a transformative technology with substantial potential to aid radiologists.
• Understanding LLM capabilities and limitations is crucial for effective integration into radiological practice.
• Future developments in LLMs promise further advancements and broader applications within radiology.