SurgicalMamba: Dual-Path SSD Model for Online Surgical Phase Recognition
Researchers have developed SurgicalMamba, a causal model for online surgical phase recognition (SPR) that addresses three key challenges in surgical video analysis: long procedures spanning tens of thousands of frames, non-uniform time flow with brief phase-defining transitions, and narrow visual domain with correlated backbone features. Built on Mamba2's structured state-space duality (SSD), SurgicalMamba holds per-frame cost at O(d) and introduces three SSD-compatible components, including a dual-path SSD block. The model is designed for context-aware operating-room systems, enabling real-time prediction from past context alone. The paper is available on arXiv under reference 2605.14889.
Key facts
- SurgicalMamba is a causal model for online surgical phase recognition.
- It is built on Mamba2's structured state-space duality (SSD).
- Per-frame cost is held at O(d).
- It addresses three demands: long procedures, non-uniform time flow, and narrow visual domain.
- A dual-path SSD block is one of its three components.
- The model enables real-time prediction from past context alone.
- The paper is available on arXiv (2605.14889).
- It targets context-aware operating-room systems.
Entities
Institutions
- arXiv