Can Gpt-4o Accurately Diagnose Trauma X-Rays? A Comparative Study with Expert Evaluations

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Radiology

Background:

• OpenAI's GPT-4o, a new AI model, can process visual data, offering potential for trauma radiography evaluation.
• Evaluating AI's role in medical diagnostics is crucial for advancing healthcare technologies.

Purpose of the Study:

• To assess the diagnostic efficacy of GPT-4o in interpreting radiographs for traumatic bone pathologies.
• To compare GPT-4o's performance against emergency medicine and orthopedic specialists.

Main Methods:

• GPT-4o and human specialists (10 emergency medicine, 10 orthopedic) interpreted 25 trauma radiograph cases.
• Cases involved traumatic bone pathologies of upper and lower extremities.
• Performance was measured by accuracy in identifying fractures or dislocations within a time limit.

Main Results:

• Human specialists achieved high accuracy (Emergency Medicine: 82.8%, Orthopedics: 87.2%).
• GPT-4o demonstrated significantly lower accuracy at 11.2%.
• Statistical analysis confirmed GPT-4o's performance was significantly inferior to specialists (p < 0.001).

Conclusions:

• GPT-4o's current radiograph interpretation for traumatic bone pathologies is limited and inferior to human experts.
• Human expertise remains essential for trauma diagnosis.
• Challenges exist in applying AI to complex medical imaging tasks.