Self-Supervised Cyclic Diffeomorphic Mapping for Soft Tissue Deformation Recovery in Robotic Surgery Scenes

Area of Science:

• Robotics
• Computer Vision
• Surgical Technology

Background:

• Recovering tissue deformation from surgical video is crucial for robotic surgery applications.
• Complex soft tissue dynamics and ambiguous pixel correspondence hinder accurate tissue tracking.
• Existing methods struggle with dense and precise deformation recovery.

Purpose of the Study:

• To develop a novel self-supervised framework for recovering dense tissue deformations from stereo surgical videos.
• To improve the accuracy and realism of tissue deformation modeling in surgical robotics.
• To address challenges posed by complex tissue manipulation and homogeneous textures.

Main Methods:

• A self-supervised framework integrating semantics, cross-frame motion flow, and long-range temporal dependencies.
• Incorporation of diffeomorphic mapping to ensure physically realistic warping fields.
• Collection and use of stereo surgical video clips from hemicolectomy and mesorectal excision procedures, covering pushing, dissection, and retraction actions.

Main Results:

• The proposed method demonstrates promising results in capturing 3D tissue deformation.
• The framework shows good generalization across different surgical actions and procedures.
• Performance surpasses current state-of-the-art methods in non-rigid registration and optical flow estimation.

Conclusions:

• This work presents the first self-supervised learning approach for dense tissue deformation modeling from stereo surgical videos.
• The developed framework offers a significant advancement in understanding and quantifying tissue dynamics during surgery.
• The method has the potential to enhance various downstream applications in robotic surgery.