UltraGS: Gaussian Splatting for Ultrasound Novel View Synthesis

The introduction of UltraGS marks a significant advancement in ultrasound imaging, a vital tool in non-invasive clinical diagnostics. Traditional ultrasound techniques face challenges due to their limited field of view, complicating novel view synthesis. UltraGS addresses this issue through a depth-aware Gaussian splatting strategy, which assigns learnable fields of view to each Gaussian, enhancing depth prediction and structural representation. Additionally, the lightweight rendering function SH-DARS integrates ultrasound-specific wave physics, accurately modeling tissue intensity. The framework is validated with the Clinical Ultrasound Examination Dataset, showcasing its superiority with state-of-the-art results in PSNR (up to 29.55), SSIM (up to 0.89), and MSE (as low as 0.002), while achieving real-time synthesis at 64.69 fps. The open-source availability of the code and dataset further promotes accessibility and innovation in the field.