LLM Report Summaries Boost Patient Comprehension, But Clinician Edits Remain Essential

In 101 outpatients, an LLM-driven web app with clinician-edited summaries, clickable terms, and AI videos significantly improved self-reported radiology report comprehension (median score 4→5, p<0.001). Yet most patients were uncomfortable with unedited AI summaries, and edits a…

• Prospective single-center study (n=101; age 20–82, mean 58±15 yrs; racially diverse) at a tertiary outpatient imaging floor; pre/post survey design, no control arm.
• LLM-generated summaries had lexical similarity of 84.63% and semantic similarity of 98.25% after clinician review, suggesting meaningful (if semantically minor) editing burden; 47.52% of participants rated LLM summaries the most helpful feature.
• Key limitation: self-reported comprehension scales (not objective testing), single institution, and no longitudinal follow-up; real-world scalability is constrained by mandatory clinician oversight adding editorial workload.

RadPigeon summaries are original and for information only. They are not clinical advice.