Chart summary LLM tool boosts radiologist efficiency

Wednesday, December 2 | 9:30 a.m.-10:30 a.m. | W3-SSIN05-7 | Room N227B

A large language model (LLM)-based chart summarization tool proved safe for clinical use and was associated with a 17% reduction in radiologists’ study review time, according to this scientific paper.

Radiologists often review large volumes of clinical documentation to understand imaging indications, which contributes to workflow inefficiency, noted presenter Saif Baig, MD, of Stanford University School of Medicine, and colleagues. The group evaluated the safety, quality, and limitations of summaries generated by Epic Imaging Insights software during a pilot deployment. They also assessed the tool’s impact on workflow efficiency.

Radiologists and trainees reviewed 102 LLM-generated summaries using a structured survey. The researchers also tracked median study review time across more than 133,000 CT and MRI exams before and after implementation in March 2026.

All of the summaries were rated safe, and most met expectations for clarity (86%) and accuracy (85%). Fewer did so for completeness (70%) and relevance (67%), and more than half had at least one issue, most commonly missing key details. Median study review time fell from 11.3 to 9.4 minutes after implementation.

“This supports LLM-based tools as a promising approach to improving radiologist efficiency,” the researchers wrote. They added that improvements in note retrieval, source attribution, and clinical context integration could further enhance the tool’s reliability.

Learn more about their experience with the LLM at this Wednesday talk.

Page 1 of 2
Next Page