Chatting with Chat GPT: Are radiation-related chats safe for mass consumption?

Purpose: We queried Chat GPT to obtain information on issues related to imaging-based radiation doses and compared the chat outputs with information on Radiology.Info (RSNA) and Radiation Protection of Patients (RPOP, IAEA) portals.

Methods/Materials: IRB approval was not required since the study did not involve any patient data use. We obtained 43 radiation-related questions on the use and effect of medical imaging-based radiation doses with radiography, CT, interventional radiology procedures, and nuclear medicine, and in children and pregnant women. The questions were obtained from both Radiology.Info (n= 18), and RPOP (n= 25) web portals. We posed the same questions to Chat GPT and a physician and radiologist compared the Chat GPT responses versus the information on the two web portals for accuracy and level and details. We recorded the word counts for each information source. Descriptive statistics were performed.

Results: Chat GPT provided answers for questions related to radiation dose safety and use matched information on both Radiology.Info (correct answers to 17/18 questions) and RPOP (correct answers to 23/25 questions) (i.e. classification of radiation-based imaging modality; pregnancy after radioiodine treatment). Chat GPT response word counts were significantly greater than those on either website for most questions (251 vs 129 words). All correct Chat GPT answers were deemed as sensible and understandable by both the physician and radiologist, but Chat GPT responses were more thorough and precise in 8/43 correct answers (i.e. on questions of radiation dosages, radiation dose quality assurance, and radioactivity).

Conclusions: For general public use, Chat GPT is a reliable source of information on radiation dose usage and safety in medical imaging. The Chat GPT provided information was sensible, understandable, and generally accurate on all aspects of radiography, CT, interventional procedures, and nuclear medicine.