Medical-AI trial: using a large language model to simplify ultrasound reports and improve patient understanding — vs. standard reports (NCT07647536)

Test reports are written in technical language so clinicians can communicate precisely with one another. But as patients increasingly read their own results online, a string of jargon easily breeds anxiety and misunderstanding. Ultrasound reports are no exception, and their meaning can be hard for patients to grasp.

This study examines whether a large language model (LLM) can rephrase reports into plain language and help patients understand.

Per the registry summary, this is a multicenter, patient-blinded controlled evaluation of whether expert-reviewed LLM-simplified ultrasound reports improve patient- or guardian-reported understanding and reading experience versus standard reports.

Routine reports were completed through existing clinical processes, and afterward participants were assigned to view either the standard report or an expert-reviewed plain-language version generated with an LLM workflow. Crucially, the simplified version is only a patient-facing communication aid; it does not replace the standard clinical report or change its content. The primary outcome is a text-comprehension composite score, and the size is 660.

What stands out is that this uses AI not for diagnosis but for health communication with patients. Patients understanding their information supports informed decisions and engagement in care. Inserting expert review and simplifying only the patient-facing aid while leaving the clinical report unchanged is a cautious, practical way to help understanding while guarding against generative-AI errors.

That said, this study evaluates understanding and reading experience; it does not imply improvement in clinical outcomes or replacement of the standard report.