Management of Burns: Multi-Center Assessment Comparing AI Models and Experienced Plastic Surgeons

Background: Burn injuries require accurate assessment for effective management, and artificial intelligence (AI) is gaining attention in burn care for diagnosis, treatment planning, and decision support. This study compares the effectiveness of AI-driven models with experienced plastic surgeons in burn assessment and management. Methods: Ten anonymized burn images of varying severity and anatomical location were selected from publicly available databases. Three AI systems (ChatGPT-4o, Claude, and Kimi AI) analyzed these images, generating clinical descriptions and management plans. Three experienced plastic surgeons reviewed the same images to establish a clinical reference standard and evaluated AI-generated recommendations using a five-point Likert scale for accuracy, relevance, and appropriateness. Statistical analyses, including Cohen's kappa coefficient, assessed inter-rater reliability and comparative accuracy. Results: AI models showed high diagnostic agreement with clinicians, with ChatGPT-4o achieving the highest Likert ratings. However, treatment recommendations varied in specificity, occasionally lacking individualized considerations. Readability scores indicated that AI-generated outputs were more comprehensible than the traditional medical literature, though some recommendations were overly simplistic. Cohen's kappa coefficient suggested moderate to high inter-rater agreement among human evaluators. Conclusions: While AI-driven models demonstrate strong diagnostic accuracy and readability, further refinements are needed to improve treatment specificity and personalization. This study highlights AI's potential as a supplementary tool in burn management while emphasizing the need for clinical oversight to ensure safe and individualized patient care.