[Large Language Models: A Comprehensive Guide for Radiologists].

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Medical Imaging

Context:

• Large language models (LLMs) demonstrate significant advancements beyond natural language processing.
• LLMs can perform domain-specific tasks, including radiology, without requiring additional fine-tuning due to extensive pre-training.
• The field of LLMs is rapidly evolving, with ongoing efforts to mitigate challenges like hallucination, data bias, high training costs, performance drift, and privacy concerns, while incorporating multimodal inputs.

Purpose:

• To provide radiologists with a comprehensive understanding of large language models.
• To offer practical guidance and an overview of the current technological landscape for LLMs in radiology.
• To explore future directions for LLM integration in medical imaging and radiological practice.

Summary:

• Contemporary LLMs, pre-trained on vast datasets, can handle diverse tasks, including specialized areas like radiology, without immediate fine-tuning.
• Emerging trends focus on small, on-premise, open-source LLMs to efficiently address medical domain knowledge, enhance privacy, and manage performance drift.
• This review synthesizes conceptual knowledge, practical advice, and a technological overview for radiologists navigating the evolving LLM landscape.

Impact:

• Empowers radiologists with knowledge of advanced AI tools for enhanced diagnostic capabilities.
• Facilitates the adoption of efficient and privacy-preserving AI solutions in clinical radiology workflows.
• Informs future research and development of LLMs tailored for the medical domain, improving patient care.