Reproduce “UltraGS: Real-Time Physically-Decoupled Gaussian Splatting for Ultrasound Novel View Synthesis” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuezhe Yang, Qingqing Ruan, Wenjie Cai, Yudang Dong, Dexin Yang, Xingbo Dong, Zhe Jin, Yong Dai

UltraGS introduces a real-time framework for sensorless B-mode ultrasound novel view synthesis, adapting Gaussian Splatting with physically-decoupled rendering and dynamic aperture rectification. The system achieves metric-consistent 3D structural reconstruction and faithful intensity synthesis from unconstrained freehand scans, demonstrating superior quantitative performance and real-time inference at 64.69 frames per second.

Key insights

  • Integrating lightweight, differentiable acoustic physics approximations into explicit radiance field representations allows for physically faithful novel view synthesis in non-optical imaging modalities like ultrasound.
  • Dynamically learning and rectifying intrinsic camera parameters (e.g., focal lengths) is crucial for maintaining geometric consistency and correcting pose drift in sensorless, unconstrained medical scanning environments.
  • Modeling ultrasound scenes with 2D Gaussian disks, combined with a perspective-correct wave-splat intersection, provides superior geometric consistency and depth recovery compared to 3D spherical Gaussians for tissue-probe interfaces.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.