GPU based Real Time Reconstruction of Compressed Sampling MRI

In magnetic resonance imaging, high-quality images are demanded to diagnosis and analytical purposes. During the acquisition process, it takes a long time to generate fully sampled k-space data for producing a high-quality image. But reducing the acquisition time creates insufficient data that degrades the image quality. Compressed sampling is a widely used technique that allows faster acquisition by undersampling k-space measurements, and it is possible to reconstruct the high-quality image from these measurements using the nonlinear algorithm. However, existing non-linear algorithms incur a long delay in the reconstruction process. In this work, bivariate Gaussian mixture model-based reconstruction algorithm is proposed to properly reconstruct the MRI image in which Graphical processing unit enabled code is developed to accelerate the reconstruction process. Experimental results show the improved performance of proposed algorithm in terms of quality of the reconstructed image and GPU-enabled methods exhibit real-time performance in terms of reducing time up to 26x, 32x, and 30x more speedups over CPU based techniques for total variation, soft-thresholding, and the proposed algorithms respectively.