AI Models Surgical Team Dynamics via Interaction Graphs

Researchers propose a real-time AI system using time-expanded interaction graphs to model surgical team dynamics. Team members are represented as time-indexed nodes with communication exchanges as directed edges, enabling dynamic interaction modeling with a static graph neural network. The system predicts procedural efficiency as deviation from expected duration and supports real-time deployment. A counterfactual analysis identifies minimal changes in communication structure for interpretable behavioral insights. The approach addresses the gap in current surgical AI, which focuses on visual workflow signals rather than structured representations of intraoperative team interactions over time.