Adaptive clutter filter design for micro-ultrasound color flow imaging of small blood vessels

In micro-ultrasound, which uses imaging frequencies above 20 MHz, obtaining color flow images (CFI) of small blood vessels using is not a trivial task because it is more challenging to suppress tissue clutter properly given the stronger blood signal power at high imaging frequencies and the slow blood velocity inside the microcirculation. To improve clutter suppression in micro-ultrasound CFI, this paper presents an adaptive clutter filtering approach that is based on a two-stage eigen-analysis of slow-time ensemble characteristics. The approach first identifies tissue pixels in the imaging view by examining whether high-frequency contents are absent in the principal slow-time eigen-components for each pixel as computed from single-ensemble eigen-decomposition. It then computes the filtered slow-time ensemble for each pixel by finding the least-squares projection residual between the pixel's slow-time ensemble and the clutter eigen-components estimated from a multi-ensemble eigen-decomposition of tissue slow-time ensembles within a spatial window. In this filtering approach, the clutter eigen-components are chosen based on whether their mean frequency lies within a spectral band. To analyze the efficacy of the proposed adaptive filter, both in-vitro experiments and Field II simulations were carried out. For the experiments, raw CFI data were acquired using a 64-element, 33 MHz linear array prototype (pulse duration: 2 cycles, PRF: 1 kHz, transmit focus: 8mm, F-number: 5). Their imaging view corresponded to the cross-section of a 0.9mm-diameter tube that was placed on top of an unsuspended table where ambient vibrations may appear; flow velocity (5, 7, 10, 15 mm/s) within the tube was controlled using a syringe pump. For the simulations, raw CFI data was computed for both plug and parabolic flow profiles, and tissue motion was modeled as 0.5 mm/s sinusoidal vibrations. For all flow velocities tested in our in-vitro study, the proposed adaptive filter improved the flow detection sensitivity as compared to existing ones. In the slow-flow case (5 mm/s), we observed over 70% increase in flow detection sensitivity (assuming a 5% false alarm rate). This effectively reduced flashing artifacts in the resulting CFIs and gave a more consistent visualization of the flow tube.