ProtoFlow: interpretable and robust surgical workflow modeling with learned dynamic scene graph prototypes - Takeaways - MDSpire

ProtoFlow: interpretable and robust surgical workflow modeling with learned dynamic scene graph prototypes

  • By

  • Felix Holm

  • Ghazal Ghazaei

  • Nassir Navab

  • May 28, 2026

  • 0 min

Share

  • 1

    ProtoFlow is a prototype learning framework designed for surgical workflow modeling using dynamic scene graphs.

  • 2

    The framework enhances interpretability and generalizability by identifying recurring structured representations of surgical interactions.

  • 3

    ProtoFlow integrates self-supervised and supervised learning, enabling effective few-shot learning in surgical video analysis.

  • 4

    Evaluation on the CAT-SG dataset demonstrates that ProtoFlow outperforms existing baselines in surgical workflow recognition.

  • 5

    The approach supports qualitative deviation analysis, providing interpretable insights into variations in surgical workflows.

Original Source(s)

Related Content