Translating radiology with AI: GPT-4o puts multilingualism to the test
Published on 3 September 2025
We keep a close eye on advances in medical AI.
A recent study published in the European Journal of Radiology evaluates GPT-4o’s performance in translating 100 radiology reports (X-rays, ultrasounds, scans, MRIs) from German into English, French, Spanish and Russian.
Key results
- High overall quality: median score of 4.5/5.
- Latin languages in the lead: English, French and Spanish surpass Russian in legibility, accuracy and usefulness.
- Fast processing time: less than 25 seconds per report.
- Factual accuracy: 84% for English, 83% for French, but only 69% for Russian.
- Potentially dangerous errors: 4% of translations, mainly in Russian.
GPT-4o shows great potential
To facilitate real-time medical translation, but requires human validation, especially for languages less well represented in the training data.
Please note: these are pure translations only. The reports are not adapted to country-specific editorial conventions. For example, the French report does not follow the classic Indication Technique Résultats Conclusion structure, but is presented in an English or German format.
In a context of medical tourism and growing linguistic diversity, the integration of LLMs into hospital systems could transform clinical communication.
