A spatiotemporal structural-feature non-local means denoising approach for contrast-free ultrasound microvascular...

Contrast-free ultrasound microvascular imaging (cf-UMI) suffers from substantial residual noise after clutter filtering due to the inherently low signal-to-noise ratio associated with unfocused plane-wave acquisition, which significantly limits the visualization of weak microvascular flow signal. Although non-local means (NLM) denoising has proven effective, its clinical applicability remains limited by prohibitive computational complexity and sensitivity to intensity variations induced by time-gain compensation (TGC). To address these limitations, we propose a spatiotemporal structural-feature non-local means (st-SFNLM) framework for efficient and robust denoising. First, st-SFNLM introduces an efficient structural-feature-based similarity evaluation to replace pixel-wise patch distance computations in NLM, enabling reduced computational complexity and improved robustness to depth-dependent intensity variations. Second, st-SFNLM performs denoising in a spatiotemporal domain rather than a purely spatial domain, exploiting the temporal continuity of blood flow signal in contrast to the temporal randomness of noise to achieve enhanced denoising performance. Comprehensive validation using flow phantom experiments at velocities of 0.2-5 mm/s and in vivo rat brain imaging demonstrates that st-SFNLM achieves enhanced microvascular contrast with superior denoising performance under low-SNR and deep-region conditions. In addition, st-SFNLM reduces computation time by 12-110× compared with conventional NLM methods, thereby facilitating its practical use in clinical cf-UMI.