Large language model-based information extraction from free-text radiology reports: a scoping review protocol.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Radiology reports contain valuable free-text data currently underutilized for secondary purposes like research.
• Information extraction (IE) using natural language processing (NLP) can unlock this data.
• Large language models (LLMs) represent a significant advancement in NLP, improving IE performance.

Purpose of the Study:

• To conduct a scoping review on the state of research for IE from free-text radiology reports using LLMs.
• To investigate the applied methods in LLM-based IE for radiology.
• To identify open challenges and limitations in current LLM-based approaches for future research guidance.

Main Methods:

• Protocol designed following the JBI Manual for Evidence Synthesis, chapter 11.2.
• Inclusion criteria and a comprehensive search strategy across four major databases (PubMed, IEEE Xplore, Web of Science, ACM Digital Library) are defined.
• Detailed description of the screening process, data charting, analysis, and presentation of extracted data.

Main Results:

• This section will be populated upon completion of the review.
• The review aims to synthesize findings on LLM applications in radiology report IE.
• Expected to identify trends, common methodologies, and gaps in the current literature.

Conclusions:

• This protocol outlines a scoping review to address the lack of comprehensive overviews on LLM-based IE in radiology.
• The findings will guide future research by highlighting effective methods and areas needing further investigation.
• The results will be published in an open-access journal focused on biomedical informatics/digital health.