Introduction


Currently, there are no (deformable) navigation systems for visceral surgery. Rigid registration is not sufficient for navigation in highly non-rigid scenes. Therefore, we introduce a method using an initial rigid registration and intraoperative monocular laparoscopic video to deform a preoperative CT during surgery.
Method Image

Method


We initialize 3D Gaussians by rigging them onto the registered mesh, extracted from the preoperative CT. Given the first camera pose, we identify the visible Gaussians, derive the visible vertices, and subsample them to initialize a sparse deformation field. For all consecutive frames, we optimize for the mesh vertices, the deformation vectors, and the appearance properties of the 3D Gaussians. Deformations are propagated to the CT.
Method Overview Image

Results


Top left is the monocular input video from the laparoscope, top right are the 4D Gaussians. Bottom shows the deforming CT.

Miscellaneous



Bibtex


@inproceedings{fehrentz2025bridgesplat,
  title={BridgeSplat: Bidirectionally Coupled CT and Non-rigid Gaussian Splatting for Deformable Intraoperative Surgical Navigation},
  author={Fehrentz, Maximilian and Winkler, Alexander and Heiliger, Thomas and Haouchine, Nazim and Heiliger, Christian and Navab, Nassir},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={44--53},
  year={2025},
  organization={Springer}
}


Acknowledgements