Dynamic mode decomposition as a framework for denoising ultrafast power doppler images

Imaging the morphology and hemodynamics of microvessels is critically important for the diagnosis and monitoring of various pathologies. Ultrafast Power Doppler (UPD) ultrasound is an emerging imaging modality for this purpose, offering a unique combination of portability, non-invasiveness, high temporal resolution, and real-time capability. However, UPD relies on unfocused wave transmission, which introduces high levels of uncorrelated noise compared to conventional Doppler imaging.

We introduce a novel denoising approach for UPD imaging based on Dynamic Mode Decomposition (DMD), a data-driven algorithm originally developed for the analysis of spatiotemporal patterns in fluid dynamics. Using a new framework that links dynamic modes to ultrasound acquisitions, temporal signals corresponding to noisy modes are removed from ultrasound data prior to the calculation of the final UPD image. Based on an energy criterion, the number of discarded modes is adapted at the pixel level, resulting in local noise filtering. The method operates after beamforming and clutter filtering, making it compatible with standard ultrafast ultrasound pipelines, and requires only a single energy-thresholding parameter.

We validated the DMD-based denoising method through simulations, phantom studies, and in vivo experiments. Compared to standard UPD images, our approach improved the signal-to-noise ratio by up to 26.0 dB and the contrast-to-noise ratio by up to 15.6 dB in vivo.

These results demonstrate that our DMD-based framework significantly enhances UPD image quality, enabling improved visualization of vessels. Beyond denoising, this method provides a principled foundation for advanced dynamic analysis in vascular ultrasound imaging.