Introducing Artificial Intelligence into Surgical Practice for Cancer Care: Clinical Applications and Ethical Considerations

Background/Objectives: Large language models (LLMs) are increasingly used in healthcare, but their utility in plastic surgery and skin cancer management remains in contention. Furthermore, the comparative performance of LLMs in the setting of skin cancer surgery is unclear. This study assessed four widely used LLMs and comparatively evaluated their relevance, accuracy, and clinical representativeness. Methods: This prospective study evaluated six standardised skin cancer prompts submitted to ChatGPT, Gemini, Claude, and Copilot. Specialist plastic surgeons assessed textual responses for relevance and accuracy, and generated images for clinical representativeness and accuracy. Readability and DISCERN scores were also assessed. Results: All four LLMs produced highly relevant textual responses, with relevance scores ranging from 4 to 5, while accuracy scores ranged from 3 to 5. ChatGPT and Copilot achieved the highest individual accuracy scores. Only ChatGPT and Gemini generated images. ChatGPT demonstrated greater clinical representativeness and accuracy across all image prompts. ChatGPT produced the most linguistically complex responses, while Copilot produced the most readable outputs. All models generated text requiring at least a high-school level of education. Claude achieved the highest mean DISCERN score (43.67), followed by ChatGPT (40.33), Copilot (36.00), and Gemini (35.67). Conclusions: In this exploratory comparison, contemporary LLM-based platforms produced relevant and largely accurate skin cancer surgery information, but differed in image generation, readability, and information quality. The generated images depicted predominantly lighter skin tones, which warrants greater input from more diverse patient populations, and outputs should be used alongside clinician review. Larger studies using patient-specific scenarios and diverse skin tones are needed.