Super-resolution velocity estimation in microvessels using Multiple Hypothesis Tracking

Imaging of microvasculature and blood flow in microvessels is a hard task for modalities like CT, MRI and Ultrasound. The image resolution (~50-200 μm) of high frequency (>20 MHz) ultrasound transducer arrays is not sufficient to image the smallest vessels (φ~8 μm). In our approach, single microbubbles' (φ~1-5 μm) positions are detected from B-mode images by background/foreground separation in super-resolution. For velocity estimation, first the problem of associating those positions to microbubbles' tracks has to be solved. The underlying combinatorial problem is addressed by the Multiple Hypothesis Tracking algorithm, which is well known in radar literature. We implemented the version, which uses Murty's K-best ranked linear assignment algorithm in order to reduce the computational burden. In a simulation experiment microbubbles were flowing through a vessel tree at a velocity of 1 mm/s. The center vessels were separated by a lateral distance of 100 μm and B-mode images were reconstructed with an ultrasound transducer array with a lateral resolution of 148 μm. The vessel could not be reconstructed with the maximum intensity method. With our approach, the vessels were clearly distinguishable and the velocity was reconstructed in super-resolution. The mean±std. blood flow velocity was calculated from microbubbles' velocities as 1.04 ± 0.07 mm/s.