A Novel Method for Ultrasound Localization Microscopy Based on Temporal Correlation Through Deep Learning

Ultrasound localization microscopy (ULM) overcomes the acoustic diffraction limit by localizing tiny microbubbles (MBs), thus enabling the microvascular to be rendered at sub-wavelength resolution. However, current techniques (both traditional and deep learning-based methods) for ULM imaging are limited to processing data from the US frame plane (x-y) for tasks (e.g., localization and tracking), with potential information along the acquisition time t-direction (between adjacent frames) remaining unexplored. In this paper, we propose a novel ULM method based on temporal correlation, termed as TC-ULM. Specifically, we exploit the potential correlations inherent in the y-t plane to further reconstruct the ULM image by extracting the variation features of MBs in the t-direction through a deep learning framework. Numerical simulation results and in vivo experiments indicate that MBs have certain continuity and local correlation between adjacent frames in the y-t plane, thus enabling the reconstruction of high-quality ULM images. At the same time, we also found that compared to previous x-y plane processing methods, TC-ULM can reduce the number of frames required for ULM imaging without sacrificing spatial resolution. This implies that CT-ULM, a novel method, can accelerate ULM imaging to some extent.