Performance benchmarking of microbubble-localization algorithms for ultrasound localization microscopy.

Ultrafast ultrasound localization microscopy can be used to detect the subwavelength acoustic scattering of intravenously injected microbubbles to obtain haemodynamic maps of the vasculature of animals and humans. The quality of the haemodynamic maps depends on signal-to-noise ratios and on the algorithms used for the localization of the microbubbles and the rendering of their trajectories. Here we report the results of benchmarking of the performance of seven microbubble-localization algorithms. We used metrics for localization errors, localization success rates, processing times and a measure of the reprojection of the localization of the microbubbles on the original beamformed grid. We combined eleven metrics into an overall score and tested the algorithms in three simulated microcirculation datasets, and in angiography datasets of the brain of a live rat after craniotomy, an excised rat kidney and a mammary tumour in a live mouse. The algorithms, metrics and datasets, which we have made openly available at https://github.com/AChavignon/PALA and https://doi.org/10.5281/zenodo.4343435 , will facilitate the identification or generation of optimal microbubble-localization algorithms for specific applications.
(© 2021. The Author(s), under exclusive licence to Springer Nature Limited.)