Minimum-Variance Ultrasonic Beamforming Implemented on High-Performance Desktop and Embedded GPUs

Abstract

Ultrasonic imaging is a widely-used medical diagnostic imaging technique, which is commonly used to observe heart movements, blood flow and fetal developments. The widespread imaging algorithm for ultrasonic imaging is delayand- sum beamforming algorithm. This algorithm is easy to implement and fast to achieve real-time performance. However, its image quality is not high enough for some complicated diagnostic scenarios. In these cases, advanced algorithms with higher image quality is required. As such, we studied minimum-variance ultrasonic beamforming algorithm which can improve ultrasonic imaging quality, and implemented it on both high-performance desktop and embedded GPUs. By applying our GPU implementation scheme, the desktop or embedded GPU implementation performance was more than 80x better than its CPU or ARM processor counterpart.