CT–Ultrasound bone surface registration · MICCAI 2026

Shape-Aware Registration Without Pretraining: CT–Ultrasound Bone Surface Alignment via Instance-Level Shape Completion

Luohong Wu1, Elise Cho1,2, Yunke Ao1,3, Lennard Marx1, Nicola Cavalcanti1, Yiru Yang4, Matthias Seibold1, Philipp Fürnstahl1

  1. 1Research in Orthopedic Computer Science, Balgrist University Hospital, University of Zurich, Zurich, Switzerland
  2. 2Computer Science Department, ETH Zurich, Zurich, Switzerland
  3. 3AI Center, ETH Zurich, Zurich, Switzerland
  4. 4University of Zurich, Zurich, Switzerland

Correspondence: luohong.wu@balgrist.ch

Code Cite

Abstract

Accurate preoperative–intraoperative registration is crucial for transferring patient-specific surgical plans to the intraoperative setting in computer- and robot-assisted orthopedic surgery. Ultrasound (US) has gained increasing attention as an intraoperative modality due to its radiation-free and real-time capabilities. However, the inherent imaging constraints, such as bone shadowing and the limited anatomical coverage during surgery, often result in incomplete US-derived bone surfaces, compromising registration accuracy and robustness. Moreover, the scarcity of paired CT–US datasets restricts the applicability of powerful learning-based approaches to different anatomies. To address these challenges, we propose ComReg, an end-to-end framework for CT–US bone surface registration via instance-specific shape completion, jointly optimizing surface completion and cross-modal registration within a unified network. In addition, we introduce a self-supervised registration strategy based on patient-specific US bone surface simulation to alleviate data scarcity, enabling effective learning of shape priors without requiring cross-instance datasets. Experiments on the publicly-available UltraBones100k dataset demonstrate the effectiveness of our approach, achieving performance comparable to state-of-the-art fully supervised methods trained on cross-instance datasets. Overall, our work provides a promising solution for cross-modality bone surface registration in limited-data settings.

Keywords: Bone surface registration · Ultrasound · Data synthesis · Bone shape completion

Figure 1 · Method overview

How ComReg works

Synthetic ultrasound is ray-cast from the patient's CT to train the model; at inference a real partial US scan is completed and aligned back to the CT — shown with the model's actual fibula output. Drag to rotate; click a step to jump.

CT bone Probe US plane Ultrasound Completed synthetic US generation
drag to rotate

Data: real ComReg output. The CT bone, partial US, completed surface, and final alignment are the model's actual output (UltraBones100k). Note: the preoperative step illustrates the ray-cast synthetic-US generation; the intra-op misalignment is representative (the per-scan disturbance is random).

Citation

Cite this work

Availability. The code is available now on GitHub. The peer-reviewed manuscript will be openly available before the conference: per the MICCAI 2026 camera-ready guidelines, the open-access proceedings are released no earlier than 21 September 2026 — roughly one week ahead of the conference (27 September – 1 October 2026). The BibTeX below will be updated with the final page numbers and DOI once the proceedings are published. 
@inproceedings{wu2026comreg,
  title     = {Shape-Aware Registration Without Pretraining: CT--Ultrasound Bone
               Surface Alignment via Instance-Level Shape Completion},
  author    = {Wu, Luohong and Cho, Elise and Ao, Yunke and Marx, Lennard and
               Cavalcanti, Nicola and Yang, Yiru and Seibold, Matthias and
               F\"urnstahl, Philipp},
  booktitle = {Medical Image Computing and Computer Assisted Intervention (MICCAI)},
  year      = {2026}
}