Unveiling Subwavelength Vascular Detail using Empirical Mode Decomposition for Super-Resolution Ultrasound Imaging

This study intends to examine the potential resolution limit for in vivo ultrasound (US) imaging, particularly focusing on super-resolution US (SRUS) for improved vascular imaging. Unlike conventional approaches that rely on specialized scanners for ultrafast frame rates, a standard clinical US machine (Acuson Sequoia 512 by Siemens Healthcare) equipped with a linear array probe operating at a modest 15 frames per sec was used for this study. Nonlinear harmonic imaging was employed at a low mechanical index to minimize the destruction of microbubbles (MBs). After a slow administration of MB contrast agents into rodents developing breast cancer, the tumor was imaged using US for a duration of $10 \mathrm{~min}$. A novel processing pipeline was developed to enhance image resolution. Images affected by breathing motion were filtered out by employing a curve fit-based technique. An empirical mode decomposition (EMD)-based spatiotemporal filter was then applied for MB detection. This was further complemented by a spatiotemporal nonlocal mean filtering (NLMF) to suppress the background noise while preserving MB tracks. Following this, MBs were localized. A bipartite graph like association approach, alongside persistence control was implemented to additionally improve the localized MB tracking fidelity. Finally, a SRUS image was constructed by projecting all the accumulated MB localizations. SRUS images depicted the detailed tumor microvascular network. The full width half maximum of microvessel cross-sectional profiles were measured and vessel as small as $43.9 \mu \mathrm{m}$ were resolved, which substantially below the diffraction limit and theoretical spatial resolution of about $220 \mu \mathrm{m}$. Results demonstrate the potential of the EMD-based SRUS technique to provide robust in vivo microvascular imaging using standard US equipment.