Multidimensional evaluation of large language models in radiology report readability - PubMed
3 hours ago
- #Large Language Models
- #Radiology Reports
- #Health Literacy
- This study evaluated the impact of demographic factors on radiology report readability and compared large language models (LLMs) in generating patient-friendly reports.
- Using a two-stage design, it involved retrospective review of 320 reports and clinical validation with 800 patients.
- All three tested LLMs significantly improved readability, with DeepSeek-R1 showing potentially superior performance.
- Demographic analysis found that higher education and older age (within similar education levels) were linked to better comprehension.
- Clinical validation indicated that simplified reports enhanced patients' subjective and objective understanding and reduced medical anxiety.
- Limitations include inconsistent model outputs, missing anatomical details, and comprehension variations due to demographics.
- The study suggests LLMs should serve as auxiliary tools for radiologists, with personalized approaches for different demographic groups.