A self-supervised framework for robust bone surface registration using 3D point clouds. It achieves high accuracy across diverse modalities (CT, ultrasound, RGB-D) and anatomies without requiring inter-subject training data.
Contour-based representations that improve registration robustness, enabling precise alignment between intraoperative X-rays and 3D anatomical models. The approach accelerates surgical planning and enhances guidance quality in orthopaedic workflows.
A curated dataset that captures the complexity of mobile noise patterns alongside strong baseline models optimized for on-device deployment. The work sets a new foundation for realistic benchmarking and rapid experimentation in mobile imaging pipelines.