SC-NeRF: Structure-constrained Neural Radiance Field for Enhanced Bone Ultrasound 3D Reconstruction

Bone’s strong attenuation leads to acoustic shadowing and imaging blind zones, while its irregular geometry results in multi-angle, nonlinear scanning trajectories, making 3D ultrasound imaging of bone highly challenging. Neural Radiance Fields (NeRF) in natural image processing leverage neural networks to simulate light propagation in space with obstacle, enabling accurate multi-view predictions and 3D reconstruction. Given the similar physics of acoustic and optical waves, this approach has been adapted for ultrasound imaging, but current methods insufficiently model the complexity of acoustic physics in bone, such as multiple reflections and scatterings, causing structural distortions and unwanted artifacts. To address this, a structure-constrained NeRF (SC-NeRF) with refined physics modeling was proposed to achieve enhanced 3D bone reconstruction.