A Bayesian Framework For Reconstruction Of Accelerated MRI Using Graph Cuts

A statistical interpretation of existing parallel magnetic resonance imaging methods reveals that simple least squares or Tikhonov regularization methods are unable to completely remove aliasing and noise from parallel MRI data during reconstruction. We present a Bayesian framework called EPIGRAM which overcomes these problems by introducing powerful edge-preserving spatial coherence priors. We show examples of in vivo reconstructed data which demonstrate significant improvement in SNR, aliasing and root mean square error properties of reconstructed images. Some extensions of the proposed method are suggested and preliminary examples are presented.