Parallel implementation of randomized singular value decomposition and randomized spatial downsampling for real time ultrafast microvessel imaging on a multi-core CPUs architecture

Singular value decomposition (SVD)-based clutter filter is widely used for functional ultrasound imaging such as super-resolution ultrasound localized microscopy and microvessel imaging. However, the demand of high computational complexity becomes one of the critical issues for the SVD-based clutter filter. Recently, randomized SVD-based (rSVD) clutter filter has been proposed for clutter suppression; the computational time can be dramatically reduced when combing with randomized spatial downsampling. In this study, we propose the use of multi-core CPU architecture to implement the rSVD-based clutter filter with randomized spatial downsampling to demonstrate that it can perform in real time. The proposed multi-core CPU architecture was embedded as an external function on a Verasonics Vantage system (Verasonics Inc., Kirkland, WA, USA). As the number of ensembles and the rank of tissue subspace are set as 50 and 20, respectively, the corresponding processing time required only around 20 ms. In addition, we also demonstrated the feasibility of real time microvessel perfusion imaging by rSVD-based clutter filter and random spatial downsampling.