Overcome limitations in computer-assisted orthopedic surgery

Ultrasound bone imaging is essential but challenging due to noise and incomplete labels, limiting progress in computer-assisted orthopedic surgery. As part of our Innosuisse Flagship project PROFICIENCY, Luohong Wu et al. at Balgrist University Hospital and Computer Vision and Geometry Group at ETH Zurich introduce a powerful new approach using CT-based automatic labeling to create UltraBones100k, the largest annotated ultrasound bone dataset to date.

Figure: Diagram explaining the transformation chain across all components. Solid black arrows represent known transformations, which are directly measured by the optical tracking system. The dashed-line arrow represents the unknown transformation from the image plane to the ultrasound probe, which will be determined through calibration.

Our model, trained on this dataset, outperforms expert’s manual labeling—even in hard-to-interpret regions—setting a new benchmark for accuracy and completeness.

This advancement paves the way for transformative breakthroughs in surgical navigation, 3D reconstruction, and multi-modal imaging integration.

Learn more details about this study here.