Leveraging Large Language Models to Enhance Radiology Report Readability: A Systematic Review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Radiology Patient Communication

Background:

• Patients have increasing direct access to their medical records, including radiology reports.
• Radiology reports are often complex and challenging for patients to comprehend.
• Large language models (LLMs) offer a potential solution for translating these reports into patient-friendly language.

Purpose of the Study:

• To systematically review the current literature on the application of LLMs for simplifying patient radiology reports.
• To identify and propose best practice guidelines for future research in this domain.

Main Methods:

• A systematic literature review was conducted following PRISMA guidelines, searching PubMed, Scopus, and Google Scholar up to February 2025.
• Studies focusing on LLM-based simplification of radiology reports for patients and evaluating readability were included.
• The Mixed Methods Appraisal tool 2018 was used for bias assessment, with findings categorized qualitatively and quantitatively.

Main Results:

• Out of 2,126 identified citations, 17 studies were included in the qualitative analysis.
• 71% of studies utilized a single LLM, with ChatGPT, Google Bard/Gemini, and Claude being prevalent.
• Quantitative readability metrics showed improvements in all assessed studies (n=12), though qualitative assessments yielded varied results based on rater type (physician vs. non-physician).

Conclusions:

• LLMs show significant potential in enhancing the accessibility and understandability of radiology reports for patients.
• Current research exhibits heterogeneity in input data, LLM models, and evaluation metrics, limiting direct comparisons.
• Establishing standardized best practice guidelines is crucial for advancing future LLM research in radiology report simplification.