A Deep Learning Method for Reduction of Microbubble Accumulation Time in Ultrasound Localization Microscopy

Ultrasound localization microscopy (ULM) enables unprecedented subwavelength spatial resolution by localizing the ultrasound contrast microbubbles (MBs). In conventional ULM methods, low-concentration MB solutions are used to ensure that the MBs are sufficiently sparse to enable localization of individual MB. This leads to a long data-acquisition time (i.e., microbubble-accumulation time) to accumulate enough MBs events using plane wave (PW) imaging. Considering the continuity of the microvessels and redundancy of MB tracks, in this study, we present a method that leverages deep neural network (DNN) to predict microvasculature using significantly reduced data-acquisition time and achieve quality comparable to that obtained using full data-acquisition time. Phantom experiments show that the proposed method significantly improves the density of localized MBs compared with the conventional method. Besides, in vivo experiments demonstrate that the proposed method is capable of reconstructing ULM images using reduced data-acquisition time with small difference, compared with the conventional method.