A singular value filter for rejection of stationary artifact in medical ultrasound

摘要

A singular value filter (SVF) is proposed for rejection of stationary clutter artifact in medical ultrasound. The SVF approach operates by projecting the original data, consisting of ensembles of complex echo data, onto a new set of bases determined from principal component analysis (PCA) using singular value decomposition (SVD). The efficacy of SVF is based on the principle that a stationary clutter signal, with perfect correlation through ensemble length, can be characterized by only the first PCA basis function, whereas significant energy contribution in the secondary PCA basis functions is necessary to describe motion and decorrelation attributed to underlying tissue structures. In contrast to many other PCA-based filtering approaches, SVF determines filter coefficients adaptively from the singular value spectrum of the original data. It is demonstrated that complex echo data is critical to the efficacy of SVF as it provides singular values that exhibit a monotonic relationship with motion complexity, and thus, provide a good means of identifying local regions of clutter. SVF is compared to a separate PCA-based technique, referred to as the blind source separation (BSS) method, as well as a frequency-based finite impulse response (FIR) clutter filter. Performance is quantified in simulated lesion images and SVF is applied to experimental mouse heart imaging data acquired from a Vevo2100 scanner (VisualSonics, Toronto, Canada) at approximately 30MHz center frequency. In simulation with levels of echo correlation expected in mouse heart imaging (0.70 correlation coefficient), SVF provided superior performance (CNR = 4.5dB) over the standard B-mode image (CNR = 2.3dB), BSS-filtered image (CNR = 3.9dB), and FIR-filtered image (CNR = 3.1dB). When SVF was applied to echo data from mouse heart images, stationary artifacts were reduced or eliminated, which enabled myocardium displacement estimates of the underlying tissue structures.