Generative Adversarial Nets for Ultrafast Ultrasound Localization Microscopy Reconstruction

Abstract

Ultrafast ultrasound localization microscopy (u ULM) can be used to resolve deep vasculature down to a few micrometers. After microbubble localization over hundreds of thousands of images, accurate and efficient tracking of each individual microbubble over consecutive frames is one of the crucial issues for uULM reconstruction. Continuous long acquisition still limits its clinical application. In the study, a generative adversarial nets (GAN) based deep learning approach is developed to facilitate microbubble tracking and further reduce the acquisition time of uULM.