Ultrasound image denoising via spatial-frequency collaborative learning

Due to inherent limitations of the imaging mechanism, ultrasound images are typically contaminated with strong noise, which not only degrades the perceptibility of critical anatomical structures but also hinders the performance of computer-aided diagnostic tasks. Existing denoising methods primarily focus on local texture or frequency modeling, yet they often struggle to simultaneously capture long-range contextual dependencies and recover fine-grained structural details. This imbalance leads to a performance bottleneck between effective denoising and structural preservation. To address this challenge, we propose MRWNet, a spatial-frequency co-enhanced denoising network, which integrates spatial retention mechanisms with frequency-domain feature modeling to improve both structural fidelity and semantic consistency. Extensive experiments demonstrate that MRWNet achieves superior performance over state-of-the-art methods in terms of PSNR, SSIM, and qualitative visual quality, validating its robustness and generalization capability under complex noise conditions.