Photo Credit: iicheck
The following is a summary of “Can GPT-4o Accurately Diagnose Trauma X-rays? A Comparative Study with Expert Evaluations,” published in the January 2025 issue of Emergency Medicine by ÖZTÜRK et al.
OpenAI’s GPT-4o, capable of processing visual data, offers new possibilities for radiographic analysis in trauma care.
Researchers conducted a retrospective study to evaluate GPT-4o’s accuracy in interpreting traumatic bone pathologies on radiographs compared to emergency medicine and orthopedic specialists.
They included 10 emergency medicine specialists, 10 orthopedic specialists, and the GPT-4o artificial intelligence (AI) model to evaluate 25 cases of traumatic bone pathologies (upper and lower extremities) from the Radiopaedia website. Participants identified fractures or dislocations on radiographs within 45 minutes, while GPT-4o completed the task in 10 distinct chat sessions.
The results showed that emergency medicine specialists achieved an average accuracy of 82.8%, while orthopedic specialists demonstrated 87.2% accuracy in interpreting radiographs. In comparison, the GPT-4o AI model reached only 11.2% accuracy. Statistical analysis indicated significant differences between the 3 groups (P < 0.001), with GPT-4o performing substantially worse than both specialist groups.
Investigators concluded the GPT-4o’s current capacity for interpreting radiographs of traumatic bone pathologies falls short of human specialists, highlighting the continued importance of human expertise in trauma diagnosis and the inherent challenges of applying AI to intricate medical imaging tasks.
Source: jem-journal.com/article/S0736-4679(24)00403-7/abstract
