Real-time GPU-based adaptive beamformer for high quality ultrasound imaging

Abstract

A real-time adaptive minimum variance (MV) beam-former realized using graphics processing units (GPUs) is presented. MV adaptive beamforming technique is attractive as it is capable of producing high quality images with narrow mainlobe width and low sidelobe level. However, because of its substantially higher computational requirements, realizing MV in real-time has been prohibitively difficult. Recent advancements in commodity GPUs have made very high performance computing possible at very affordable price. Using a commercial off-the-shelf GPU, an MV beam-former achieving real-time performance has been realized. Tradeoffs between computational throughput and image quality have been studied. Careful selection of algorithm parameters, including receive aperture and sub-aperture size, was demonstrated to be imperative for achieving real-time performance without sacrificing image qualities.