Performance of Foreground-Background Separation Algorithms for the Detection of Microbubbles in Super-Resolution Imaging

Ultrasound localization microscopy is a super- resolution imaging technique using ultrasonic contrast agents: To visualize microvasculatures, the injected microbubbles (MB) are localized and tracked. For the detailed reconstruction of the vessel trees, an accurate localization of the MB as well as a compensation of their positions regarding tissue motion are fundamental. In clinical practice, different motion artifacts can be present: pulsatile heart beating, breathing, and patient or transducer displacements, resulting in periodic and aperiodic motion patterns. To enable both, the detection of MB as well as the estimation of tissue motion without disturbances due to moving MB, the B-mode sequences are usually separated into MB (foreground)and tissue (background)images. This separation can be carried out with different methods. Here, the performances of the Singular Value Decomposition (SVD), spatiotemporal non-local-means filtering (stNLM) and rank filtering are investigated and compared regarding the detection reliability and the accuracy of motion estimation. We found substantial differences in the performances depending on the measurement scenario and the required evaluation after separation (detection or motion estimation). Therefore, a combination of two methods, e.g. spatiotemporal NLM for MB detection and rank filtering for motion estimation should be used.