A Regularized MRI Image Reconstruction based on Hessian Penalty Term on CPU/GPU Systems

In this paper we investigate an inverse reconstruction problem of Magnetic Resonance Imaging with few acquired body scanner samples. The missing information in the Fourier domain causes image artefacts, therefore iterative computationally expensive recovery techniques are needed. We propose a regularization approach based on second order derivative of both simulated and real images with highly undersampled data, obtaining a good reconstruction accuracy. Moreover, an accelerated regularization algorithm, by using a projection technique combined with an implementation on Graphics Processing Unit (GPU) computing environment, is presented. The numerical experiments give clinically-feasible reconstruction runtimes with an increase in speed and accuracy of the MRI dataset reconstructions.