Pilot Study on AI Image Analysis for Lower-Limb Reconstruction-Assessing ChatGPT-4's Recommendations in Comparison to Board-Certified Plastic Surgeons and Resident Physicians.

Abstract

AI, especially ChatGPT, is impacting healthcare through applications in research, patient communication, and training. To our knowledge, this is the first study to examine ChatGPT-4's ability to analyze images of lower leg defects and assesses its understanding of complex case reports in comparison to the performance of board-certified surgeons and residents. We conducted a cross-sectional survey in Switzerland, Germany, and Austria, where 52 participants reviewed images depicting lower leg defects within fictitious patient profiles and selected the optimal reconstruction techniques. The questionnaire included cases with varied difficulty, and answer options did not always include the most obvious choices. Findings highlight that ChatGPT-4 successfully evaluated various reconstruction methods but struggled to determine the optimal solution based on the available information in visual and written forms. A chi-squared test of independence was performed to investigate the overall association between answer options (A, B, C, and D) and rater group (board-certified surgeons, ChatGPT-4, and resident). Inter-group rater associations showed significant overall test results (p < 0.001), with high agreement among board-certified surgeons. Our results suggest that board-certified plastic surgeons remain essential for patient-specific treatment planning, while AI can support decision-making. This reaffirms the role of AI as a supportive tool, rather than a replacement, in reconstructive surgery.