Elith CTO Koki Inoue’s paper, “Assessing knowledge about medical physics in language-generative AI with large language model: using the medical physicist exam,” has been accepted in the journal Radiological Physics and Technology.
We will continue to pursue research and development aimed at further advancing AI technology.
■ About the accepted co-authored paper
This research evaluates how accurately the large language models ChatGPT-3.5 and ChatGPT-4.0 can answer Japan’s medical physicist certification exam, aiming specifically to benchmark their knowledge in the field of medical physics. The results showed that while ChatGPT-4.0’s accuracy rate was high overall, it fell below 60% in certain categories (such as radiation measurement and related regulations), demonstrating that the findings can serve as foundational data for future medical physics tool development.
Paper link
https://link.springer.com/article/10.1007/s12194-024-00838-2