Unsupervised Semantic Constrained Denoising Framework for Ultrasound Plane-Wave Imaging

Ultrasound plane-wave (PW) imaging achieves ultrahigh frame rates but suffers from strong diffraction artifacts. The standard reconstruction approach consists of compounding successive transmissions coherently, at the expense of the frame rate. Deep learning-based ultrasound imaging has emerged as a promising avenue to improve the quality of PW images. However, existing research predominantly focuses on supervised learning that relies on high-quality paired References, which imposes challenges in practical use. In this study, we introduce a novel unsupervised semantic-consistent denoising framework (SECOND) for PW imaging. The framework integrates a generative learning framework with contrastive learning, which effectively reduces diffraction artifacts in PW images and preserves structural details in reconstructed images. We evaluated the proposed approach with both numerical simulation and real acquisition. Experimental results demonstrate that the proposed approach yields superior qualitative and quantitative performance compared to existing learning-based works, and competes with the standard compounding of multiple PW transmissions.