Cheetah: A Library for Parallel Ultrasound Beamforming in Multi-Core Systems
David Romero-LaordenCarlos Julian Martin-ArguedasJavier Villazon-TerrazasOscar Martinez-GraulleraMatilde Santos PenasCesar Gutierrez-FernandezAna Jimenez Martin
摘要：Developing new imaging methods needs to establish some proofs of concept before implementing them on real-time scenarios. Nowadays, the high computational power reached by multi-core CPUs and GPUs have driven the development of software-based beamformers. Taking this into account, a library for the fast generation of ultrasound images is presented. It is based on Synthetic Aperture Imaging Techniques（SAFT） and it is fast because of the use of parallel computing techniques. Any kind of transducers as well as SAFT techniques can be defined although it includes some pre-built SAFT methods like 2R-SAFT and TFM. Furthermore, 2D and 3D imaging（slicebased or full volume computation） is supported along with the ability to generate both rectangular and angular images. For interpolation, linear and polynomial schemes can be chosen. The versatility of the library is ensured by interfacing it to Matlab, Python and any programming language over different operating systems. On a standard PC equipped with a single NVIDIA Quadro 4000（256 cores）, the library is able to calculate 262,144 pixels in ≈105 ms using a linear transducer with 64 elements, and 2,097,152 voxels in ≈ 5 seconds using a matrix transducer with 121 elements when TFM is applied.
2015 Workshop 8
基础科学; 工程科技Ⅱ辑; 信息科技
物理学; 工业通用技术及设备; 计算机软件及计算机应用