SurgStreamer: proactive, request-driven monitoring of surgical video streams with structured state and a reusable reasoning layer
SurgStreamer formalizes request-driven event detection for surgical video by separating procedure-specific perception from procedure-agnostic reasoning. It maps raw video to a structured state of tools and actions, allowing a reusable reasoning core to monitor specific conditions. The system achieved a macro event-F1 of 0.598, proving the separation improves performance over existing streaming vision models.