Identifying Volar Locking Plates on Plain Radiographs: Can Artificial Intelligence Models 'Beat' Clinicians?

Area of Science:

• Orthopaedic Surgery
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of volar locking plates on plain radiographs is crucial for hand surgeons, particularly for revision surgeries, implant removal, or managing periprosthetic fractures.
• Challenges in implant identification arise when surgical history is unknown or originates from different institutions or countries.
• The potential of artificial intelligence (AI) in medical image recognition is recognized, but its efficacy in orthopaedic implant identification requires thorough investigation.

Purpose of the Study:

• To compare the performance of an openly available AI model, ChatGPT 5, against experienced hand consultants in identifying volar locking plates on radiographs.
• To assess the accuracy of AI and human experts in distinguishing between implants from various manufacturers.

Main Methods:

• Fifty-two radiographs featuring distal radius plates from 10 major implant manufacturers were collected from open-access sources.
• An AI program (ChatGPT 5) and five hand consultants independently attempted to identify the manufacturer of each plate.
• Accuracy rates were calculated for each participant, with statistical comparisons made between the AI and consultants using McNemar's test and logistic regression.

Main Results:

• ChatGPT 5 achieved a low accuracy of 5.8% (3 out of 52 radiographs).
• Hand consultant accuracies varied from 13.5% to 46.2%, with a mean accuracy of 30.8%.
• Four out of five consultants significantly outperformed the AI (p < 0.01), and the human group collectively showed over seven times higher odds of correct identification compared to the AI (OR 7.26, p < 0.001).

Conclusions:

• Identifying volar locking distal radius plates from plain radiographs remains a significant challenge, even for experienced surgeons, with top performers achieving less than 50% accuracy.
• Current non-specialized AI tools like ChatGPT 5 are not suitable for clinical orthopaedic implant identification.
• Future development of specialized AI models trained on curated orthopaedic datasets may offer potential solutions for improving implant identification accuracy in clinical practice.