Contrast-free super-resolution ultrasound imaging based on two-stage diffusion probabilistic models

Super-resolution ultrasound (SR-US) imaging techniques, also known as ultrasound localization microscopy (ULM), substantially improves the spatial resolution of ultrasound (US) imaging and is helpful for microvessel imaging. However, their clinical applications face challenges due to the reliance on ultrasound contrast agents (UCAs), calling for contrast-free SR-US imaging techniques. Several research groups have utilized endogenous red blood cells (RBCs) for contrast-free imaging, yet the improvement of spatial resolution remains limited. In this study, we introduce a contrast-free SR-US imaging framework based on a two-stage diffusion probabilistic model, termed as 2S-DUS, to overcome this limitation. The first stage employs an efficient diffusion model to transform low-resolution power Doppler (PD) images into super-resolution maps, while the second stage refines vascular continuity and anatomical fidelity using a physics-guided enhancement module. Validated on in vivo contrast-free rat brain datasets, 2S-DUS achieves a spatial resolution of ∼24.6 μm (full-width at half-maximum, FWHM), successfully performing SR-US imaging. Quantitative evaluations demonstrate its superior performance in perceptual metrics (LPIPS: 0.234; DISTS: 0.154) and resolution metrics compared to state-of-the-art methods (RS, ULM-GAN, ULM-MbCNRT). Ablation studies confirm the advantages of the two-stage architecture, balancing generative realism and structural fidelity. The findings of this work highlight the effectiveness of the proposed 2S-DUS in realizing contrast-free SR-US imaging, offering transformative potential for clinical applications in cerebrovascular disease monitoring and diagnosis.