Informatics in Medicine Unlocked · Published 2026-02-05 · DOI 10.1016/j.imu.2026.101746
Artificial intelligence (AI) methods for surgical step recognition have shown promising results, but their ability to generalise across different types of surgery remains a critical challenge. This study aimed to assess whether an AI algorithm designed for step prediction in videos of plate osteosynthesis for distal radius fractures could be directly applied to fibula osteosynthesis.The algorithm was originally developed to recognise six steps of distal radius osteosynthesis: installation, approach, distal fixation, proximal fixation, verification, and closure. A fibula osteosynthesis video was recorded and annotated with six steps: installation, approach, fracture reduction, plate fixation, verification, and closure. As fracture reduction is a procedure-specific step not included in the original training set, algorithm predictions were analysed with and without this step. Predictions were also evaluated at different frame rates (5 and 1 fps).The results showed that the algorithm achieved better performance on distal radius osteosynthesis videos, with accuracy, precision, and recall exceeding 80%, whereas performance remained below 80% for fibula osteosynthesis. The best predictions were obtained at 1 fps when the reduction step was excluded, highlighting both the influence of temporal resolution and the limitations of cross-procedure generalisation. Differences between the two surgeries studied are discussed to explain the observed prediction errors.In conclusion, while the algorithm can be applied to fibula osteosynthesis as a proof of concept, its reduced performance indicates that procedure-specific fine-tuning is necessary. These findings suggest that the algorithm could serve as a foundation for the development of AI algorithms for other types of osteosynthesis.
Abstract from DOAJ. Public domain (CC0 1.0).
Read the article at the publisher →