3D Ultrasound Microbubbles Localization Using Object Detection Model

Ultrasound Localization Microscopy (ULM) is a cutting-edge imaging technique that unveils the intricate microvascular structures within organs. This method relies on the dynamic properties of microbubbles as contrast agents combined with sophisticated post-processing algorithms to achieve high-resolution imaging. Despite its success in 2D super-resolution medical imaging, the ULM faces significant challenges when applied in 3D ULM, including trade-offs between accuracy, time consumption, and the handling of complex data with substantial memory requirements. Addressing these challenges necessitates innovative approaches to improve the efficiency and accuracy of ULM analysis. In this work, we proposed a 3D object detection convolutional neural network to solve MBs localization, which outputs the coordinates of the predicated MBs and their corresponding confidence scores. The precision accuracy can reach up to 2.6 pixels, and the Jaccard index is up to 91%. In addition, the processing time for each frame was 0.003 s.