GPU accelerated MIMO ultrasonic imaging using propagation and back-propagation method

In this paper, we develop efficient hardware implementation strategies and methods for wave based multiple input multiple output ultrasonic tomographic imaging using Graphic Processing Units. By exploiting the underlying parallelism of the iterative image reconstruction algorithm, we test different Compute Unified Device Architecture (CUDA) kernel configurations in order to measure the changes in the processing time of our algorithm. By examining the imaging speed and the resulting image quality, we develop an optimal kernel configuration that maximizes the throughput of the CUDA implementation for the ultrasonic imaging method. The results show an improvement by a factor of 80 when compared to the C/C++ version of the algorithm, and by a factor of 515 when compared to the MATLAB version while achieving high quality tomographic images.