Large Language Model Helps Radiology Department Boost Patients’ Report Comprehension

Why It Matters

The study shows that large language models can materially improve patient understanding of complex imaging results, a key driver of engagement and outcomes, but scaling requires clinician validation to address safety and trust concerns.

Key Takeaways

• GPT‑5 generated summaries raised patient comprehension scores to perfect 5/5
• 48% of participants rated AI summaries as most helpful tool
• Clinician edits removed ~25 words per summary, ensuring accuracy
• Only 54% felt comfortable with AI summaries without physician oversight
• Web app added clickable terms and AI‑generated videos for education

Pulse Analysis

Large language models are rapidly moving beyond research labs into frontline clinical tools, and radiology is a natural entry point because imaging reports are dense with technical jargon. By converting raw reports into concise, lay‑person summaries and pairing them with interactive elements, GPT‑5 helps bridge the communication gap that often leaves patients confused about their diagnoses. The Emory pilot demonstrates that AI can act as a first‑line educator, delivering consistent explanations at scale while freeing clinicians to focus on nuanced discussions.

The prospective trial enrolled 100 outpatients who accessed their post‑scan reports through a custom web portal. Each report was processed by GPT‑5 using a standardized prompt, then reviewed by board‑certified radiologists who trimmed an average of 24.75 words to eliminate hallucinations. Patients completed a five‑point comprehension survey, showing a median jump from 4 to 5 after viewing the AI‑augmented content. Notably, 48% of respondents singled out the AI‑generated summary as the most useful feature, underscoring the value of concise, digestible language. However, only about half of participants felt fully comfortable with unsupervised AI output, highlighting lingering trust issues.

The broader implication for health systems is clear: AI‑enhanced reporting can boost patient engagement, potentially improving adherence to follow‑up care and satisfaction scores. Yet successful deployment hinges on integrating clinician oversight into the workflow, a factor that adds labor and may slow adoption. Regulatory bodies will likely scrutinize the provenance of AI‑generated content, demanding transparency and audit trails. As vendors refine prompting techniques and develop automated quality‑checks, the balance between efficiency and safety will dictate how quickly such tools become a standard component of patient‑centered radiology services.