Home | Publications | YNF25

MoViS: Motion-Guided Video Generation for Laparoscopic Surgery

MCML Authors

Yousef Yeganeh

→ Group Nassir Navab
Computer Aided Medical Procedures & Augmented Reality

Nassir Navab

Prof. Dr.

Principal Investigator

Computer Aided Medical Procedures & Augmented Reality

Azade Farshad

Dr.

→ Group Nassir Navab
Computer Aided Medical Procedures & Augmented Reality

Abstract

Realistic surgical video synthesis is vital for simulation, training, and preoperative planning, yet existing methods often fall short in capturing the intricate motions intrinsic to surgical procedures. In this paper, we introduce a novel framework that generates high-fidelity surgical videos by integrating keypoints extracted via a zero-shot keypoint prediction module with the motion guidance of video diffusion models. Our approach decouples spatial content from temporal dynamics: critical keypoints representing anatomical landmarks and instrument positions are first identified in a zero-shot manner and then used to steer the diffusion process, ensuring that generated video frames adhere to plausible and consistent motion patterns. Extensive experiments on curated surgical video datasets demonstrate that our method produces temporally coherent videos with enhanced motion accuracy and visual realism compared to baseline models. The results suggest that our framework can substantially contribute to surgical training and simulation by generating dynamic and realistic surgical scenarios that faithfully reflect complex procedural movements.

inproceedings

CREATE @MICCAI 2025

Workshop on Clinical-Driven Robotics and Embodied AI Technology at the 28th International Conference on Medical Image Computing and Computer Assisted Intervention. Daejeon, Republic of Korea, Sep 23-27, 2025.