Reproduce “DBT-Bleed: Dual-Branch Temporal Modeling with Key-Frame Selection for Surgical Bleeding Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sudhanshu Mishra, Jialang Xu, Jensen Ang, Evangelos B. Mazomenos, Beng Ti Ang, Yueming Jin
The DBT-Bleed framework, featuring dual-branch multi-scale temporal modeling and a hierarchical entropy-driven key-frame selection strategy, improves the detection of surgical bleeding as Intraoperative Adverse Events. The system achieves superior performance on the MultiBypass dataset and demonstrates robust zero-shot generalization across different surgical procedures.
Key insights
- Multi-scale temporal modeling effectively distinguishes active bleeding IAEs from residual blood by leveraging both short- and long-term progression patterns across different semantic levels.
- A hierarchical entropy-driven key-frame selection strategy efficiently mitigates video redundancy while retaining critical temporal information for IAE detection.
- A CLIP-based dual-branch architecture, augmented with spatial and temporal adapters, enables robust zero-shot generalization of bleeding IAE detection across diverse surgical procedures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.