SpainHealthcare and life sciences2025 – ongoing

Hospital Universitario Morales Meseguer

GPT-4o assistant for structured on-call radiology reports

A Murcia hospital tested a GPT-4o assistant for radiology residents' on-call reports across 480 reports

In progressIndependent study

Problem

Residents on emergency radiology shifts write reports under time pressure, and the wording varies from one doctor to the next. Clear, uniform reports matter to the physicians who requested the scans.

Approach

Radiologists at the hospital built a custom GPT-4o assistant for structured reporting. Four residents wrote 480 reports, alternating between writing by hand and writing with the assistant. The team compared them with 200 anonymised reports by attending physicians.

Results

480reports written by four residents, alternating manual and LLM-assistedIndependent study[1]

4.75-4.90 of 5median Likert scores across six dimensions reported in the studyIndependent study[1]

Takeaway

The hospital used the assistant to standardise reports sent to doctors who requested scans. The pilot involved four residents.

Sources

Pilot study of an LLM-based structured reporting system in emergency radiology (García Hidalgo et al.)
Spanish Journal of Medical Education (Revista Española de Educación Médica)19 Jan 2026Research paper or report

Source summaries

[1]Spanish Journal of Medical Education (Revista Española de Educación Médica)19 Jan 2026

Researchers at Hospital Morales Meseguer tested a custom GPT-4o assistant for structured on-call radiology reporting. Four residents wrote 480 reports, and 200 attending reports served as a comparison. Median scores were 4.75 to 4.90 out of 5 across six dimensions, and residents' own reports were told apart from assisted ones with an F1 score of 0.92.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026