Wendy Sun

While there are many upsides to patients having immediate access to their test results—often seeing them before their clinical team does—this can leave them confused or anxious when the results are difficult to interpret.

In hopes of making test results more accessible, a research team led by Mass General Brigham’s Wendy Sun and Omar Arnaout, MD, analyzed 500 neuroradiology reports, including brain MRIs, spine MRIs and head CTs. A large language model (GPT-4o) was prompted to simplify each report to a middle-school reading level and translate some of them into Spanish while preserving clinical accuracy.

Two neuroradiologists then independently evaluated the AI-generated summaries for any omissions, hallucinations or imprecisions, while certified medical interpreters assessed the quality of the translations.

Omar Arnaout, MD

They found the AI-generated reports were easier to read, conveyed findings in a neutral and less fearful tone and had low rates of clinically relevant errors. In blinded evaluations, the AI’s Spanish translations were not significantly different from translations completed by human experts.

While these findings show AI’s potential promise in improving the readability and multilingual accessibility of medical reports, the authors emphasize that clinician oversight and additional validation remain of utmost importance as these tools move toward clinical implementation.

Published in Scientific Reports on May 21, 2026| Read the paper: “Patient-friendly simplification and translation of neuroradiology impressions using artificial intelligence”

Summary reviewed by: Wendy Sun, lead author; Omar Arnaout, MD, senior author