Decoding the Surgical Scene: A Scoping Review of Scene Graphs in Surgery
A scoping review of 52 studies on scene graphs in surgery reveals a significant increase in research, with a notable shift towards foundation and generative AI models. However, a 'data divide' persists, with most research using real-world endoscopic video while external operating room modeling relies on simulations. The review proposes a new 'Validation Trinity' framework to address the gap between current computer vision metrics and the needs for clinical validation of these neuro-symbolic AI systems. AI
IMPACT Proposes a new validation framework for neuro-symbolic AI in surgery, aiming to bridge the gap to clinical practice.