Harnessing Large Language Models for Radiology Report Simplification and Improving Patient Comprehension: A Narrative...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Radiology Reporting

Background:

• Radiology reports use technical language, hindering patient comprehension.
• Digital imaging advancements haven't resolved patient understanding barriers.
• Large language models (LLMs) offer potential for simplifying reports.

Purpose of the Study:

• To review LLMs for simplifying patient-centered radiology reports.
• To assess the effectiveness and challenges of LLM-based report simplification.
• To explore implications for clinical practice and radiologist workflow.

Main Methods:

• Narrative review of 19 studies on LLMs (GPT-3.5, GPT-4, Claude, Gemini).
• Analysis of studies evaluating LLM performance across various imaging modalities.
• Examination of readability improvements and accuracy concerns.

Main Results:

• LLMs significantly improved report readability, lowering reading levels from 10th-14th to 5th-8th grade.
• Studies reported accuracy issues, including omissions, commissions, and distortions.
• Variability in accuracy was noted based on imaging modality and specific LLM used.

Conclusions:

• LLMs show transformative potential for enhancing patient understanding of radiological findings.
• Ensuring accuracy and clinical precision during simplification remains a significant challenge.
• Careful implementation with oversight is crucial for effective LLM integration into radiology.