ArthroPhase: A Novel Dataset and Method for Phase Recognition in Arthroscopic Video

The ArthroPhase dataset comprises 27 full-length videos of Anterior Cruciate Ligament (ACL) reconstruction surgeries, each meticulously annotated with five key surgical phases: Preparation, Diagnosis, Femoral Tunnel Creation, Tibial Tunnel Creation, and ACL Reconstruction. Designed specifically for arthroscopic procedures, the dataset captures the unique challenges of this domain, including limited field of view, instrument occlusions, and visual distortions caused by irrigation fluids and surgical debris. ArthroPhase serves as a benchmark resource for advancing research in surgical workflow analysis, computer-assisted interventions, and automated phase recognition in arthroscopic surgery. This work also received support from OR-X, a Swiss national research infrastructure for translational surgery, with funding provided by the University of Zurich and the University Hospital Balgrist.