Quantitative Ultrasound Localization Microscopy by Attention‐Guided Deep Learning Method

ABSTRACT Ultrasound localization microscopy (ULM) achieves subwavelength resolution by localizing individual microbubbles (MBs), yet current methods mainly emphasize positional accuracy while neglecting MB amplitude information, which carries valuable physiological significance. This study proposes a deep learning (DL)‐based quantitative ULM framework, termed qULM‐AGDL, to jointly recover MB position and amplitude. The network integrates a multi‐scale fusion channel–spatial attention block within an end‐to‐end convolutional neural network (CNN) to enhance feature learning for both localization and quantification. Simulation and experimental results demonstrate that qULM‐AGDL achieves high localization accuracy (8.98 ± 4.83 µm) and low amplitude error (33.89 ± 16.75) even under overlapping or low‐signal‐to‐noise ratio (SNR) conditions. Compared with the mSPCN‐ULM and modified Gaussian fitting methods, qULM‐AGDL provides significantly improved performance ( p < 0.001) in terms of mean squared error (MSE) and structural similarity (SSIM). The proposed framework establishes a basis for quantitative ULM and holds potential for functional and clinical applications such as perfusion assessment and vascular monitoring.