SA-NeRF: Scattering-aware Neural Radiance Fields for Robust 3D Ultrasound Imaging

Clinically, 3D ultrasound imaging is often limited by surrounding anatomical structures, such as ribs and superficial organs, which obstruct views and degrade image quality. Accurately reconstructing 3D structures from 2D slices acquired at varying angles and irregular paths remains challenging. In computer vision, Neural Radiance Fields (NeRF) employ neural networks to simulate light propagation, enabling high-quality reconstruction of occluded views. Given the similar propagation properties of acoustic and light waves, this physics-driven deep learning method has been applied to ultrasound 3D reconstruction. However, inadequate modeling of complex acoustic propagation leads to blurred and distorted tissue structures. To address this, we propose a scattering-aware NeRF (SA-NeRF) guided by precise physical modeling for high-fidelity 3D reconstruction.