Ultrasound Localization Microscopy Learned from power doppler by uncertainty frequency density estimation and semantic consistency awareness.

Authors

Abstract

Ultrasound Localization Microscopy (ULM) achieves micron-level vascular visualization beyond the resolution of conventional ultrasound imaging by tracking microbubble positions. However, ULM relies on high-frame-count ultrasound images, which leads to long acquisition times, poor real-time performance, and high post-processing costs. Therefore, we propose a Power Doppler (PD) to ULM image super-resolution method named PDSR. Specifically, to address the issues of content distortion and training instability in PD image super-resolution, we propose a Patch Consistency Regularization (PCR), which enhances the representation capability of unpaired image translation models through cross-region patch-wise information interaction. Then, we propose a Semantic Consistency Awareness (SCA) to constrain the semantic alignment between PD and ULM images in high-weight regions sampled under discriminator guidance, reducing the generator's tendency to translate false structures from PD images. Finally, we propose an uncertainty frequency Density Variation Constraint (DVC), which enhances vessel realism by constraining the translation of high-information-density regions in PD images to corresponding regions in ULM images. Extensive experiments and ablation studies demonstrate that the proposed method achieves state-of-the-art performance in PD-to-ULM image translation, attaining an SSIM of 78.45% and a PSNR of 15.03   dB, while requiring only 13.1 ms per image for reconstruction. Given its unpaired training strategy and high-fidelity imaging results in preclinical rat studies, PDSR offers potential for enabling contrast-free ULM imaging. Code is available at: https://github.com/LQH89757/PDSR.

Tags

Topics