Optimization of real-time ultrasound PCIe data streaming and OpenCL processing for SAFT imaging

Abstract

Our goal is to develop a complete ultrasound platform based on real-time SAFT (Synthetic Aperture Focusing Technique) GPU processing. We are planning to integrate all the ultrasound modules and processing resources (GPU) in a single rack enclosure with the PCIe switch fabric backplane. The first developed module (RX64) provides acquisition and streaming of 64 ultrasound channels. We implemented and benchmarked data streaming from the RX64 to the GPU memory and the SAFT image reconstruction on the GPU. A high system performance was achieved using hardware assisted direct memory transfers and pipelined processing workflow. The complete system throughput, including 128 channel data transfer at 16kS per line and low-resolution 256×256 pixel image SAFT reconstruction on a single Nvidia K5000 GPU, reached 450 fps. The obtained results proved the feasibility of the ultrasound real-time imaging system with GPU SAFT processing.