Back to 2026 ePosters
Utility of Open Source Artificial Intelligence in the Identification and Surgical Management of Hand Fractures
Tiffany Shi, PhD
1, Maria Rich, BS
1, Ermina Lee, BS
1, Trent James, BA
1, Uzair Qazi, MD
2; Ann R Schwentker, MD
3(1)University of Cincinnati College of Medicine, Cincinnati, OH, (2)University of Cincinnati College of Medicine, CINCINNATI, OH, (3)Cincinnati Childrens Hospital and Medical Center, Cincinnati, OH
Background: Open source artificial intelligence (OSAI) has been shown to have exciting potential in the field of plastic surgery to aid in the identification and management of fractures. Previous studies have assessed OSAI's ability to identify osteologic conditions of the hand; however, OSAI's ability to determine surgical necessity and create an operative plan for identified fractures is still unclear. In this study, we evaluated the ability of ChatGPT to both correctly identify and manage hand fractures.
Methods: 111 cases, filtered to include only those with x-ray identified fractures of the wrist and hand, were pulled from Radiopaedia, an open-source radiology database. Case images were uploaded to ChatGPT (v4.0), and each case was queried for a specific diagnosis, whether the injury required operative management, and, if yes, a step-by-step breakdown of the surgical procedure. For cases where ChatGPT correctly identified the fracture, hand surgeons were given the same radiologic imaging and provided their own assessment on whether the injury required operative management.
Results: ChatGPT correctly identified 37/111 (33.3%) hand and wrist fracture cases pulled from Radiopaedia. Of these 37 cases, ChatGPT identified 20 (54.1%) as requiring surgical management. When further broken down into cases with only one fracture present, ChatGPT identified 11/26 (42.3%) as requiring surgical management. In radiographs with multiple fractures present, ChatGPT determined 9/11 (81.8%) cases as requiring surgical management. ChatGPT was significantly better at correctly identifying fractures in adult versus pediatric cases (p<0.05). However, there was no significant difference between the frequency of cases identified as needing surgical management. Of cases where ChatGPT correctly identified fractures, hand surgeons agreed with 27/37 (73.7%) of ChatGPT's surgical management decisions. No significant difference (p=0.61) was seen between frequency of metacarpal (9/13, 69.2%), phalangeal (5/9, 55.5%), distal radius (11/14, 78.6%), and wrist (1/1, 100%) fractures where at least one hand surgeon agreed with ChatGPT's determination of requirement for surgical fixation. Based solely on imaging, agreement in management between hand surgeons occurred in 26/37 (70.3%) of cases.
Conclusions: While it was significantly more effective in correctly identifying adult hand fractures, ChatGPT was still broadly error-prone, with limited ability to identify hand fractures when provided x-ray imaging. In cases where fractures were correctly identified, ChatGPT was largely able to correctly identify if surgical management was warranted. Further enhancement of OSAI is necessary to warrant its reliable usage as a tool to aid surgeons in the identification and management of surgical cases.
Back to 2026 ePosters
