Related Experiment Video
Updated: May 22, 2026
09:41
A Pipeline for 3D Multimodality Image Integration and Computer-assisted Planning in Epilepsy Surgery
Published on: May 20, 2016
Endo-PairGS: pair priors for dynamic endoscopic scene reconstruction
Xiankang Yu1, Yuichiro Hayashi2, Masahiro Oda2,3
1Graduate School of Informatics, Nagoya University, Furo-cho, Chikusaku, Nagoya, Aichi, Japan. yu.xiankang.n0@s.mail.nagoya-u.ac.jp.
Summary
Endo-PairGS improves dynamic endoscopic scene reconstruction by using aligned point cloud pairs for better initialization of Gaussian splatting (GS) models. This method enhances 3D scene completeness, especially in challenging surgical scenarios.
Area of Science:
- Computer Vision
- Medical Imaging
- Robotics
Background:
- Dynamic scene reconstruction is vital for minimally invasive surgery and surgical navigation.
- Gaussian splatting (GS) shows promise for reconstructing dynamic endoscopic scenes.
- Challenges include deformable tissues, occlusions, and changing camera views, hindering effective GS model initialization.
Purpose of the Study:
- To propose Endo-PairGS, a novel framework for initializing 4D Gaussian splatting models for dynamic endoscopic scene reconstruction.
- To leverage aligned point cloud pairs from different frames to overcome initialization challenges.
Main Methods:
- The framework involves static and dynamic 3D reconstruction.
- A foundation model is fine-tuned on endoscopic data to generate aligned point clouds, using optical flow masks for stability with deformable tissues.
- Paired point clouds are used to initialize a 4D GS model, complementing occluded or unseen regions.
Main Results:
- Endo-PairGS outperformed existing methods on EndoNeRF and StereoMIS datasets in both quantitative and qualitative evaluations.
- Achieved 32.47 PSNR and 0.871 SSIM on StereoMIS (P3), representing a 4% and 4.8% improvement over the baseline, respectively.
Conclusions:
- Endo-PairGS offers a more complete initialization method, significantly enhancing dynamic endoscopic scene reconstruction performance.
- The study provides a robust solution for challenging endoscopic scenarios.
- Code and data will be publicly released.