Super-Resolution Ultrasound Localization Microscopy via Brownian Bridge Diffusion Model for Liver Fibrosis Assessment

The hepatic vasculature exhibits a strong correlation with the progression of liver fibrosis (LF), underscoring the importance of microvascular assessment in liver disease evaluation. Ultrasound Localization Microscopy (ULM) can generate super-resolution images of liver vasculature, presenting substantial potential for LF assessment. However, ULM requires localizing and tracking microbubbles across tens of thousands of frames to achieve high-quality imaging, resulting in a lengthy acquisition and processing time. Moreover, the extended acquisition time introduces motion artifacts from respiration and cardiac activity, posing a significant challenge, especially in organs such as the liver. Power Doppler (PD) imaging reconstructs blood flow from only hundreds of frames, enabling short-time acquisition but with limited resolution and signal-to-noise ratio. In this work, we introduce a novel Brownian Bridge Diffusion Model (BBDM) that establishes a diffusion bridge between the distributions of PD and ULM images. Moreover, we evaluated our approach with LF rat models (N=62), ensuring comprehensive assessment across all fibrosis stages (F0–F4). The experimental results demonstrate that our method can generate high-quality ULM-like images, effectively accelerate ULM imaging, and have significant potential for the evaluation of liver diseases.