SurgicalMamba: Dual-Path SSD with State Regramming for Online Surgical Phase Recognition

Researchers introduced SurgicalMamba, a causal surgical video model for online phase recognition built on Mamba2 structured state-space duality.

The model uses a dual-path recurrent design, intensity-modulated stepping, and state regramming to better handle long procedures and irregular phase changes.

It reported state-of-the-art results on seven benchmarks, including Cholec80 and AutoLaparo, under online evaluation.

SurgicalMamba aims to improve real-time operating-room understanding with lower per-frame cost, supporting workflow monitoring and context-aware assistance.