2026
MemSurg: Memory-Augmented Long-Horizon Reasoning for Surgical Video Understanding
Anonymous Authors
Submitted to MICCAI 2026
MemSurg is a memory-augmented framework for long-horizon surgical video understanding that leverages guided prompting and chain-of-thought reasoning. By constructing an external surgical memory graph from segmentation and motion cues, it retrieves task-relevant evidence across time and composes more coherent prompts for downstream inference. This design improves consistency on instrument, action, and workflow understanding, outperforming GPT-4o and by 22.42%.
MemSurg: Memory-Augmented Long-Horizon Reasoning for Surgical Video Understanding
Anonymous Authors
Submitted to MICCAI 2026
MemSurg is a memory-augmented framework for long-horizon surgical video understanding that leverages guided prompting and chain-of-thought reasoning. By constructing an external surgical memory graph from segmentation and motion cues, it retrieves task-relevant evidence across time and composes more coherent prompts for downstream inference. This design improves consistency on instrument, action, and workflow understanding, outperforming GPT-4o and by 22.42%.