Up-Coming Events

活动日期： 2025年5月29日

Below are the speakers' details: 

Mr. Felix Holm

PhD Candidate, 

Chair of Computer Aided Medical Procedures (Prof. Navab), 

Technical University Munich, Germany 

Title: Holistic Surgical Video Understanding using Scene Graphs

Biography: 

Felix Holm is a PhD candidate at the Chair for Computer Aided Medical Procedures at Technical University Munich with Professor Nassir Navab. His research focuses on surgical video understanding using advanced computer vision techniques, particularly scene graph representations for modeling surgical procedures. He has published multiple works on dynamic scene graphs for surgical applications, including papers at ICCVW 2023 and MICCAI 2025. His work addresses critical challenges in surgical data science and contributes to making surgical procedures safer and more efficient through AI-powered analysis of surgical videos, with particular expertise in cataract surgery modeling and interpretable machine learning approaches for healthcare applications. 

This presentation explores the application of scene graphs for comprehensive surgical video understanding, addressing critical challenges in modern surgical data science. Scene graphs provide a structured representation that models surgical tools, anatomical structures, and their dynamic interactions throughout procedures. The talk presents three key contributions: Dynamic Scene Graph representation for temporal surgical modeling, CAT-SG - the first large-scale cataract surgery scene graph dataset with over 1.8 million annotated relations, and ProtoFlow - an interpretable workflow modeling approach using learned prototypes. These methods demonstrate significant improvements in surgical phase recognition, technique classification, and anomaly detection while maintaining robustness and data-efficiency. The holistic approach enables applications including automated report generation, intraoperative warnings, robotic assistance, and surgical education. Results show that scene graphs not only improve performance but also provide explainable insights into surgical procedures, making AI systems more transparent and trustworthy for clinical deployment.