GPU Acceleration of PROPELLER MRI Using CUDA

PROPELLER technique can effectively cancel motion artifacts in MRI. But its wider application in clinical situation is limited due to considerable reconstruction times. Since most correction operations in PROPELLER reconstruction can be done for each strip respectively, the algorithm is highly parallelizable. This allows us to exploit the data-parallel nature of the GPU fragment processors. The paper presents an implementation to accelerate reconstruction of PROPELLER MRI on graphics processing units (GPUs) using CUDA. An improved grid-driven interpolation algorithm for PROPELLER trajectory is proposed for real-time imaging applications. The experiments show that the reconstruction is speeded up about nine times on GPU that of implementation on CPU with compatible motion correction accuracy and image quality.