Power Doppler (PD) is the preferred technique for accessing weakly scattering blood cells with low-velocity flow. However, strong clutter decreases sensitivity of the blood signals, which degrades micro-vasculature imaging. Eigen-based filtering methods have been developed to separate the independent scattering sources using both spatial and slow-time statistics of echo data. We recently found that extending data into frame-time sampled on the order of Hz increases the sensitivity of the weak perfusion signal. Higher-order singular value decomposition (HOSVD) effectively decouples these sources. The challenge is to select suitable subspaces to isolate the desired blood components. We propose a subspace selection strategy for optimal HOSVD performance.