Improved Ultrasound Localization Microscopy Using Deep Learning Enhanced Beamformer

Ultrasound localization microscopy (ULM) has emerged as an effective tool to visualize the structural and hemodynamic information of microvessels with a scale of 10 μm, through localization and tracking of injected microbubbles (MBs). Due to the unfocused transmit of ultrafast plane wave imaging, high background noise is introduced. In addition, commonly used localization methods cannot balance the noise suppression and weak MB localization well. Adaptive beamformers usually utilize the signal coherence of different channels or transmit angles, improving the distinction between noise and MBs with low scattering intensity and potentially improving the performance of ULM. However, a trade-off exists between computational cost and imaging quality of adaptive beamformers, hindering their application on ULM. To accelerate the high-quality beamforming procedure of ULM from a mass of data, in this study, we propose a deep learning enhanced beamformer (Uformer-ULM) to break this trade-off. Contrast-enhanced in-vivo rat brain and rat liver experiments were conducted. Results show that the proposed method can visualize more subtle vessels in ULM velocity maps, which are invisible using conventional delay-and-sum beamformer-based ULM (DAS-ULM). In addition, it can also achieve better delineation of relatively large vessels in the velocity maps. The results demonstrate that the proposed Uformer-ULM method can facilitate high-quality ULM.