Multidimensional Clutter Filter Optimization for Ultrasonic Perfusion Imaging

Abstract

Combinations of novel pulse-echo acquisitions and clutter filtering techniques can improve the sensitivity and the specificity of power Doppler (PD) images, thus reducing the need for exogenous contrast enhancement. We acquire echoes following bursts of Doppler pulse transmissions sparsely applied in regular patterns over long durations. The goal is to increase the sensitivity of the acquisition to slow disorganized patterns of motion from the peripheral blood perfusion. To counter a concomitant increase in clutter signal power, we arrange the temporal echo acquisitions into two data-array axes, combine them with a spatial axis for the tissue region of interest, and apply 3-D singular-value decomposition (SVD) clutter filtering. Successful separation of blood echoes from other echo signal sources requires that we partition the 3-D SVD core tensor. Unfortunately, the clutter and blood subspaces do not completely uncouple in all situations, so we developed a statistical classifier that identifies the core tensor subspace dominated by tissue clutter power. This paper describes an approach to subspace partitioning as required for optimizing PD imaging of peripheral perfusion. The technique is validated using echo simulation, flow-phantom data, and in vivo data from a murine melanoma model. We find that for narrow eigen-bandwidth clutter signals, we can routinely map phantom flows and tumor perfusion signals at speeds less than 3 mL/min. The proposed method is well suited to peripheral perfusion imaging applications.

Deep Learning in Ultrasound Imaging

“…One of the most popular approaches for clutter suppression is spatio-temporal filtering based on the singular value decomposition (SVD). This strategy has led to various techniques for clutter removal [68], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80]. SVD filtering includes collecting a series of consecutive frames, stacking them as vectors in a matrix, performing SVD of the matrix and removing the largest singular values, assumed to be related to the tissue.…”

Combining Spatial Registration With Clutter Filtering for Power-Doppler Imaging in Peripheral Perfusion Applications

“…Our observations of Doppler spectra from in vivo peripheral perfusion data [ 17 ], [ 18 ] made before and after spatial registration are generally symmetric about the origin, suggesting that in vivo muscle perfusion is diffuse. Spectral symmetry is the reason for power-Doppler methods being more informative than color-flow methods in perfusion imaging applications.…”

“…However, if the tissue and blood-cell speeds overlap [ 18 ], the separation of clutter and blood singular-value subspaces is incomplete. Some investigators have proposed rigid spatial registration of echo frames before clutter filtering [ 14 ], [ 15 ], [ 16 ], [ 17 ], [ 18 ]. Registration minimizes any in-plane translation of the dominant clutter echoes between frames, which narrows the clutter singular-value bandwidth and improves the separation between clutter and blood subspaces to enhance PCA filter efficiency.…”

Clutter suppression in ultrasound: performance evaluation and review of low-rank and sparse matrix decomposition methods

“…SVD based on Tucker decomposition is generally called high-order singular value decomposition (HOSVD) [97,98], which calculates the singular values of the three expansions U 1 , U 2 , U 3 of a three-dimensional tensor under Tucker decomposition. HOSVD-based ultrasound clutter optimization has been proposed [52,99] and proved to be more robust to low sampling rates than SVD.…”

“…Therefore, plane-wave ultrasound has a higher and more urgent filtering requirement than traditional CFI. Recent methods have extended SVDbased clutter suppression to a higher order by analyzing a data tensor instead of a twodimensional matrix [42,48,52]. Since the first few singular values do not necessarily correspond to the clutter signal in the presence of a large temporal misalignment among the frames, the motion correction step has been introduced in SVD-based clutter rejection [53].…”