Norikazu Matsutomo
   Department   Kawasaki University of Medical Welfare  ,
   Position   Associate Professor with Special Assignment
Article types 原著
Language English
Peer review Peer reviewed
Title Can interactive artificial intelligence be used for patient explanations of nuclear medicine examinations in Japanese?
Journal Formal name：Annals of nuclear medicine
Abbreviation：Ann Nucl Med
ISSN code：18646433/09147187
Domestic / ForeginDomestic
Volume, Issue, Page 39(8),pp.774-780
Author and coauthor Norikazu Matsutomo, Mitsuha Fukami, Tomoaki Yamamoto
Publication date 2025/04
Summary OBJECTIVE:This study aimed to evaluate the accuracy and validity of patient explanations about nuclear medicine examinations generated in Japanese using ChatGPT- 3.5 and ChatGPT- 4.METHODS:ChatGPT was used to generate Japanese language explanations for seven single-photon emission computed tomography examinations (bone scintigraphy, brain perfusion imaging, myocardial perfusion imaging, dopamine transporter scintigraphy [DAT scintigraphy], sentinel lymph node scintigraphy, lung perfusion scintigraphy, and renal function scintigraphy) and 18F-fluorodeoxyglucose positron emission tomography. Nineteen board-certified nuclear medicine technologists evaluated the accuracy and validity of the responses using a 5-point scale.RESULTS:ChatGPT- 4 demonstrated significantly higher accuracy and validity than ChatGPT- 3.5, with 77.9% of responses rated as above average or excellent for accuracy, in comparison to 36.3% for ChatGPT- 3.5. For validity, 73.1% of ChatGPT- 4's responses were rated as above average or excellent, in comparison to 19.6% for ChatGPT- 3.5. ChatGPT- 4 outperformed ChatGPT- 3.5 in all examinations, with notable improvements in bone scintigraphy, lung perfusion scintigraphy, and DAT scintigraphy.CONCLUSION:These findings suggest that ChatGPT- 4 can be a valuable tool for providing patient explanations of nuclear medicine examinations. However, its application still requires expert supervision, and further research is needed to address potential risks and security concerns.
DOI 10.1007/s12149-025-02047-2
PMID 40234374