Graphics Processing Units Accelerated MIMO Tomographic Image Reconstruction Using Target Sparseness

摘要

GPU computing of medical imaging applications adds an extra layer of acceleration after mathematical algorithms are used to reduce computation times. Our work improves the performance of the multiple-input multiple-output ultrasonic tomography algorithm, by using target sparseness and GPUs with CUDA. The main goal was to determine how GPUs can be best used to accelerate sparsity-aware algorithms for ultrasonic tomography applications. We present smart kernels to compute portions of the algorithm that exploit GPU resources such as shared memory and computing units that can be applied to other applications. Using our accelerated algorithm, we analyze different sparsity constraints setups and evaluate how GPU ultrasonic tomography with target sparseness behaves against the same algorithm that does not incorporate prior knowledge of target sparseness.