← Back to home

Automatic Multi-View X-Ray/CT Registration Using Bone Substructure Contours

Roman Flepp1,2, Leon Nissen2, Bastian Sigrist2, Arend Nieuwland2, Nicola Cavalcanti2, Philipp Fürnstahl2, Thomas Dreher1,3, Lilian Calvet2
1 University Children's Hospital Zürich, Switzerland
2 Research in Orthopedic Computer Science (ROCS), University Hospital Balgrist, University of Zurich, Switzerland
3 Department of Orthopedic Surgery, University Hospital Balgrist, University of Zurich, Switzerland
Paper Code Dataset (DICOM/NIfTI) Dataset (Zenodo)

01. Overview

Accurate intraoperative X-ray/CT registration is essential for surgical navigation in orthopedic procedures. Existing methods struggle with consistently achieving sub-millimeter accuracy, robustness under broad initial pose estimates, or require manual key-point annotations. We propose a novel multi-view X-ray/CT registration method for intraoperative bone registration that addresses these challenges.

The method consists of a multi-view, contour-based iterative closest point (ICP) optimization. Unlike previous methods, which attempt to match bone contours across the entire silhouette in both imaging modalities, we focus on matching specific subcategories of contours corresponding to bone substructures. This reduces ambiguity in the ICP matches, resulting in a more robust and accurate registration solution that requires only two X-ray images and operates fully automatically.

  • Multi-view, contour-based ICP registration matching bone-substructure contours instead of whole-silhouette contours, reducing correspondence ambiguity.
  • Fully automatic pipeline needing only 2 X-ray images and no manual key-point annotation.
  • Sub-millimeter accuracy: 0.67mm mean reprojection distance vs. 5.35mm for a manually-assisted commercial solution (~8x improvement).
  • New public cadaveric dataset: 5 specimens with real X-ray images, contour masks, X-ray poses, and paired CT scans.
  • Accepted to IPCAI 2025, published in the International Journal of Computer Assisted Radiology and Surgery.

02. Method & Implementation

Substructure Contour Extraction: Two intraoperative X-ray images are acquired and used to extract the bone substructure contours. These contours are obtained using a semantic segmentation U-Net trained on patient-specific synthetic data, with a loss function designed to favor thin, connected structures. In this example, we extract contours of the medial and lateral condyles as well as the femoral shaft. Simultaneously, the intrinsic and extrinsic parameters of the X-ray imaging system are calibrated using a custom fiducial marker.
CT Model Reprojection: Preoperatively, the bone model is segmented from the CT scan, and its substructures are defined. Using the previously calibrated X-ray parameters, contours of the bone model and its respective substructures are projected onto the X-ray image planes.
Multi-View Substructure ICP Optimization: The extracted bone substructure contours are aligned to the projected CT model contours using a multi-view Iterative Closest Point (ICP) optimization, considering multiple views simultaneously and matching semantically extracted bone substructure contours with their corresponding projections from the CT model. For simplicity, this example shows the process with only one view. The optimization iteratively refines the alignment until convergence, yielding a robust and accurate registration by reducing matching ambiguity.
Example of Multi-View Registration: This video demonstrates the complete registration process. On the left, the setup features two calibrated X-ray images; on the right, the iterative registration process is visualized, showing the alignment of CT model substructure reprojections with the extracted bone contours. The initial pose begins with a challenging 180° rotation around the femoral shaft axis, typically difficult for standard registration methods.

03. Results

The proposed registration method is evaluated on real X-ray images using mean reprojection distance (mRPD).

Method Mean Reprojection Distance Manual Intervention
Commercial solution 5.35 mm Required
Proposed method (ours) 0.67 mm None (fully automatic)

The method consistently achieves sub-millimeter accuracy, an ~8x improvement over the commercial baseline, while offering improved practical applicability by being fully automatic. Our method offers a practical, accurate, and efficient solution for multi-view X-ray/CT registration in orthopedic surgeries, which can be easily combined with tracking systems, enhancing intraoperative navigation and contributing to more accurate surgical outcomes.

04. Citation

A machine-readable Markdown version of this page (abstract, key contributions, results, BibTeX) is available at papers/bone-substructure-contours-xray-ct-registration.md. 

@article{Flepp2025,
  author  = {Flepp, Roman and Nissen, Leon and Sigrist, Bastian and
             Nieuwland, Arend and Cavalcanti, Nicola and
             Fürnstahl, Philipp and Dreher, Thomas and Calvet, Lilian},
  title   = {Automatic Multi-View X-Ray/CT Registration Using Bone Substructure Contours},
  journal = {International Journal of Computer Assisted Radiology and Surgery},
  year    = {2025},
  month   = may,
  day     = {20},
  doi     = {10.1007/s11548-025-03391-4},
  url     = {https://link.springer.com/article/10.1007/s11548-025-03391-4},
}