Leveraging Autoencoder Neural Networks to Improve Microbubble Detection and Localization in Ultrasound Localization Microscopy

This paper introduces a novel approach designed to enhance the process of microbubble (MB) detection and localization (MBDL) within ultrasound localization microscopy (ULM). ULM serves as a tool for detecting, localizing, tracking, and visualizing intravenously injected MB contrast agents, enabling the generation of super-resolved hemodynamic maps of microvascular networks. The proposed method involves utilizing convolutional neural network (CNN) in conjunction with the drainage-divide technique (DDT) to improve the quality of MBDL, thereby enhancing the resulting ULM images. Numerical experiments conducted on a simulated dataset demonstrate the effectiveness of the proposed approach compared to the most efficient state-of-the-art method.