Use of Large Language Models on Radiology Reports: A Scoping Review.

Area of Science:

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

Background:

• Large language models (LLMs) are emerging tools with significant potential in radiology.
• Applications include enhancing clinical workflows, diagnostic accuracy, and patient communication.

Purpose of the Study:

• To conduct a scoping review of current and emerging uses of LLMs in radiology text.
• To assess LLM capabilities, limitations, and methodologic considerations in radiologic applications.

Main Methods:

• A comprehensive literature search was performed on PubMed and Embase.
• 69 relevant articles were included in the review to synthesize findings.

Main Results:

• LLMs excel in report simplification and translation but show mixed results in classification tasks.
• Fine-tuning and structured prompting enhance LLM accuracy; managing stochasticity remains a challenge.
• Most studies documented dataset independence and prompting methods, but fewer addressed LLM variability.

Conclusions:

• LLMs offer transformative potential in radiologic care, particularly in workflow optimization.
• Further research is needed to address validation, generalizability, and responsible implementation.
• This review guides stakeholders in leveraging LLMs effectively and ethically in radiology.