AI vs. MD: Benchmarking ChatGPT and Gemini for Complex Wound Management

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Wound Management

Background:

• Hard-to-heal wounds present a significant clinical and economic burden globally.
• Emerging AI tools, specifically Large Language Models (LLMs), offer potential for clinical decision support.
• The reliability of LLMs in wound care assessment requires rigorous evaluation.

Purpose of the Study:

• To assess the accuracy of ChatGPT and Gemini in visually describing complex wounds.
• To evaluate the reliability of these LLMs in recommending initial therapeutic management based on clinical images.
• To compare AI performance against expert clinical consensus.

Main Methods:

• Twenty diverse clinical wound images were analyzed by ChatGPT and Gemini.
• Standardized prompts were used to query the AI models.
• AI-generated assessments were compared to a gold standard established by three plastic surgeons.

Main Results:

• No significant difference in overall performance was found between ChatGPT, Gemini, and the expert panel.
• Gemini showed a higher agreement rate for management recommendations (75.0% vs. 60.0%).
• Both LLMs accurately identified vascular lesion etiologies and critical red flags like ischemia, but showed limitations with atypical cases.

Conclusions:

• LLMs show considerable potential as educational and decision-support tools in wound care, especially for non-specialists.
• AI can expedite diagnosis and standardize initial management for complex wounds.
• Image-based AI reasoning has limitations; professional supervision is crucial for optimal patient care pathways.