Real-time realization of adaptive dynamic quadrature demodulation on a gpu-based ultrasound imaging system

In medical ultrasound imaging, the frequency- and depth-dependent attenuation causes the degradation in signal-to-noise ratio (SNR) in quadrature demodulation (QDM). To improve SNR, the adaptive dynamic QDM (ADQDM) method based on a 2nd-order autoregressive (AR) spectral estimation was previously proposed. However, due to its high computational requirements, it is challenging to implement the ADQDM in real time. In this paper, the optimal realization of ADQDM on a GPU-based ultrasound imaging system is presented. To efficiently implement the method, the image is divided into multiple zones, and the center frequency of a receive signal at each zone is independently estimated by using the 2nd-order AR model. The estimated center frequencies are used for dynamic quadrature demodulation. This method was incorporated on the Compute Unified Device Architecture (CUDA) platform and throughputs were measured using a NVIDIA´s GTX-560Ti GPU chip. The evaluation was conducted with the beamformed 6144×256 pixel radio-frequency (RF) data which were captured by a commercial ultrasound scanner from the liver of a volunteer. The total execution time for ADQDM is 3.44 ms, which indicates that it can be implemented in real time on a GPU-based medical ultrasound system.