Low-Complexity Rank-Revealing Spatiotemporal Microvascular Clutter Filter

Effective clutter rejection filters are critical in microvascular Doppler imaging. SVD-based modeling is a successful approach to spatiotemporal decluttering, however requires adaptation to many overlapping pixel neighborhoods. Unfortunately, the SVD is compute-intensive and resistant to parallelization. Another matrix decomposition is Lanczos Bi-diagonalization (BD). Because the BD weights are ordered, termination at R iterations sets the model rank. Thus, the low-rank nature of the clutter signal permits good approximation with only partial decomposition. Processing is effective in microvascular imaging, and the bulk of computation is matrix-vector multiplication, efficiently parallelized in numeric libraries.