More Research

AutoSeg4ETICA

Deep learning analysis of the Eustachian tube and internal carotid artery.

Otolaryngology - Head and Neck Surgery 2024
AutoSeg4Sinonasal

Label-efficient automated sinonasal CT segmentation.

Otolaryngology - Head and Neck Surgery 2024
SelExNet

Joint RF-Gradient optimization for high-fidelity reduced-FOV 7T MRI excitation.

Magnetic Resonance in Medicine 2026

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

Shizhan Gong1, Yonghao Long1, Kai Chen1, Jiaqi Liu1, Yuliang Xiao1, Alexis Cheng2, Zerui Wang2, Qi Dou1
1The Chinese University of Hong Kong 2Cornerstone Robotics Limited
IEEE Transactions on Medical Imaging 2024
IEEE Paper Code

Through a novel self-supervised framework, the recovered deformations can represent actual tissue dynamics by integrating semantics, cross-frame motion flow, and long-range temporal dependencies from stereo surgical videos.

Abstract

The ability to recover tissue deformation from visual features is fundamental for many robotic surgery applications. This has been a long-standing research topic in computer vision, however, is still unsolved due to complex dynamics of soft tissues when being manipulated by surgical instruments. The ambiguous pixel correspondence caused by homogeneous texture makes achieving dense and accurate tissue tracking even more challenging. In this paper, we propose a novel self-supervised framework to recover tissue deformations from stereo surgical videos. Our approach integrates semantics, cross-frame motion flow, and long-range temporal dependencies to enable the recovered deformations to represent actual tissue dynamics. Moreover, we incorporate diffeomorphic mapping to regularize the warping field to be physically realistic. To comprehensively evaluate our method, we collected stereo surgical video clips containing three types of tissue manipulation (i.e., pushing, dissection and retraction) from two different types of surgeries (i.e., hemicolectomy and mesorectal excision). Our method has achieved impressive results in capturing deformation in 3D mesh, and generalized well across manipulations and surgeries. It also outperforms current state-of-the-art methods on non-rigid registration and optical flow estimation. To the best of our knowledge, this is the first work on self-supervised learning for dense tissue deformation modeling from stereo surgical videos.

Results

Quantitative evaluation of our method compared with existing methods. We train the model with all training data and report the soft tissue manipulation-level metric on the test set. The best number for each category is highlighted in bold.

Models %|JΦ| <= 0 ↓ l1-norm ↓ PSNR ↑ SSIM (%) ↑
push dissect retract total push dissect retract total push dissect retract total push dissect retract total
NICP 3.53 6.58 3.87 4.65 22.78 24.04 26.34 24.46 17.85 17.92 16.90 17.53 64.92 63.92 58.34 62.25
CPD 8.14 7.21 7.44 7.56 22.31 21.74 24.21 22.82 17.84 18.42 17.54 17.89 61.21 63.88 58.46 60.90
SIFT 0.15 0.14 0.41 0.24 21.03 20.85 25.31 22.53 18.78 18.92 17.65 18.41 76.46 76.64 71.47 74.73
Harris-Laplace 0.15 0.14 0.41 0.26 21.02 20.94 25.23 22.53 18.79 18.87 17.67 18.40 76.46 76.59 71.58 74.64
RAFT 3.64 3.86 3.37 3.61 13.64 12.00 13.61 12.89 21.92 21.60 22.19 21.90 84.29 85.59 82.19 83.97
UFlow 3.63 3.83 3.35 3.59 12.98 12.03 13.69 12.97 21.94 22.01 22.18 22.14 84.52 85.82 82.23 84.13
Ours (+RAFT) 0.01 0.01 0.03 0.02 13.03 11.94 13.43 12.80 22.02 23.09 21.73 22.27 84.61 85.97 82.82 84.42
Ours (+UFlow) 0.01 0.01 0.03 0.02 12.86 12.00 13.56 12.83 22.11 23.12 21.67 22.28 84.70 86.03 82.69 84.43

BibTeX

@article{gong2024self,
  title={Self-Supervised Cyclic Diffeomorphic Mapping for Soft Tissue Deformation Recovery in Robotic Surgery Scenes},
  author={Gong, Shizhan and Long, Yonghao and Chen, Kai and Liu, Jiaqi and Xiao, Yuliang and Cheng, Alexis and Wang, Zerui and Dou, Qi},
  journal={IEEE transactions on medical imaging},
  volume={43},
  number={12},
  pages={4356--4367},
  year={2024}
}