Decoding Radiology Reports: Artificial Intelligence-Large Language Models Can Improve the Readability of Hand and...

Area of Science:

• Radiology
• Medical Informatics
• Artificial Intelligence

Background:

• Radiology reports often contain complex medical terminology, hindering patient comprehension.
• Improving patient understanding of diagnostic imaging results is crucial for informed healthcare decisions.

Purpose of the Study:

• To evaluate the effectiveness of an Artificial Intelligence-Large Language Model (AI-LLM) in simplifying hand and wrist radiology reports.
• To assess the impact of AI-LLM on the readability and accuracy of these reports for patient understanding.

Main Methods:

• Extracted 300 hand and wrist radiology reports (radiographs, CT, MRI).
• Utilized an AI-LLM with a prompt to translate reports into layman's terms.
• Calculated Flesch reading ease scores (FRES) and Flesch-Kincaid reading levels (FKRL) for original and AI-generated reports.
• Assessed AI report accuracy using a Likert scale and recorded "hallucinations".

Main Results:

• AI-LLM significantly improved FRES and FKRL scores across all report types.
• AI-generated reports achieved a mean reading level below an eighth-grade level.
• Mean Likert scores for accuracy were high (radiograph: 4.1, CT: 3.9, MRI: 3.9), with low hallucination rates (3-6%).

Conclusions:

• AI-LLM technology effectively enhances the readability of hand and wrist radiology reports.
• AI-LLM presents a patient-centric strategy to improve comprehension of imaging findings.
• This technology holds promise for more accessible and understandable medical reporting.