Ultrasound Super Resolution Using Deep Learning Based on Attention Mechanism

Ultrasound Localization Microscopy (ULM) has gained a lot of interest as a new imaging technology capable of achieving subwave diffraction resolution. Currently, it is still challenging to achieve a high accuracy and robust localization in in-vivo dataset. Traditional single emitter localization methods, such as Gaussian fit, Radial Symmetry (RS) and average weight had problems with precision, robustness and computational efficiency. In this work, we propose an attention mechanism based neural network, namely ATT-net, to make an end-to-end mapping to localize the microbubbles and scale the input dimension. The performance of the proposed method is validated on in-silico and in-vivo data and compared with two other localization methods. The results showed that our proposed network achieved higher precision and Jaccard index. These benefits can be used to further improve the image visualization and processing efficiency.