Significant advancement in surgical phase recognition for arthroscopy

The study – performed by Ali Bahari Malayeri et al. at Balgrist University Hospital as part of our Innosuisse Flagship project PROFICIENCY - is introducing the first ACL-specific dataset and a powerful transformer-based model for surgical phase recognition.

Figure: Overview of the final model for surgical phase recognition, The architecture has a transformer branch which processes an 80-second sequence of features. These features are derived from our spatio-temporal feature extractor, which combines spatial information from ResNet-50 with temporal context through transformer layers. Both phase and SPI predictions are performed at the frame level to ensure temporal synchronization.

By integrating spatio-temporal features and a novel Surgical Progress Index, the model effectively handles the visual complexity of real procedures. Achieving strong performance across two datasets, it sets a new benchmark for accuracy and reliability in phase recognition. This work opens the door to smarter surgical training, real-time assistance, and greater efficiency in orthopedic surgery.

More details about this study you can learn here.