Comparison of localization methods in super-resolution imaging

Abstract

Ultrasound localization microscopy (ULM) produces vascular images with a resolution ten times better than the conventional ultrasound. Generating ULM images requires several steps: injection of the microbubbles (MBs) into the bloodstream, image acquisition, detection, localization, and tracking of MBs. Finally, the accumulation of all MBs positions produces the final image. In this context, the localization step is crucial, and it is still under discussion what is the optimum localization method. In this study, four different MBs localization methods were compared using a simulation along with experimental validation in a microtube phantom. Ultrasound interactions were simulated with the FullWave Solver in a two-dimensional domain. This solver was set to simulate 1000 frames of a plane-wave propagating in a heterogeneous tissue with different diameters of vessels, containing blood and MBs flowing with a parabolic fluid velocity profile. Four methods to localize the MB were compared: Weighted centroid, 2D spline, paraboloid fitting, and onset detection. Four metrics were used to compare the methods; MB distribution error; Full-width-at-half-maximum (FWHM) error; Number of MBs detected per frame; and computational time cost. Paraboloid fitting was the most robust method, regarding different diameters. When compared to the weighted centroid, paraboloid fitting improves the estimates of the localization profile by 56%, and detects near 100% of the MBs, while weighted centroid only detected 25% of them. The precision of the onset detection method depends on the vessel diameter, showing good results only for small vessels. Most importantly, the proposed simulations can be seen as a tool that offers access to the ground truth of many MBs’ localization under relatively realistic physics.