Real-time Adaptive and Localized Spatiotemporal Clutter Filtering for Ultrasound Small Vessel Imaging

Abstract

Effective clutter filtering is crucial in suppressing tissue clutter and extracting blood flow signal in Doppler ultrasound. Recent advances in eigen-based clutter filtering techniques have enabled ultrasound imaging of microvasculature without the need for contrast agents. However, simultaneously achieving fully adaptive, highly sensitive and real-time implementation of such eigen-based filtering techniques in clinical scanning scenarios for broad translation remains challenging. To address this, here we propose a fast spatiotemporal clutter filtering technique based on eigenvalue decomposition (EVD) and a novel localized data processing framework for robust and high-definition ultrasound imaging of blood flow. Unlike the existing local clutter filter that hard splits the ultrasound data into small blocks, our approach applies a series of 2D spatial Gaussian windows to the original data to generate local data subsets. This approach improves performance of flow detection while effectively avoiding undesired grid artifacts with dramatically reduced number of subsets required in local EVD filtering to shorten computation time. By leveraging the computational power of Graphics Processing Units (GPUs), we demonstrate the real-time implementation capability of the proposed approach. We also introduce and systematically evaluate several adaptive and automatic eigenvalue thresholding methods tailored for EVD-based filtering to facilitate optimization of blood flow imaging for either global or localized processing. The feasibility of the proposed clutter filtering technique is validated by experimental results from phantom and different in vivo studies, revealing robust clinical application potential. A tradeoff between improved performance and computational cost associated with the packet size and subset number in local processing is also investigated.