Ashish Raj1, Ramin Zabih2
1Radiology, Weill Medical College of Cornell University, New York, NY, USA; 2Computer Science, Cornell U, Ithaca, NY
Current parallel imaging methods like SENSE and GRAPPA are seriously limited by SNR and g-factors associated with coil geometry. Most MR images have sharp edges and are piecewise smooth. A new edge preserving Markov Random Field (MRF) prior model was proposed in to address this issue, and a new graph cut-based algorithm was developed to solve the computationally challenging non-convex optimization problem. Here we propose two important extensions of the EPIGRAM method by (a) incorporating 3-D spatial priors and by jointly reconstructing multi-slice images, and (b) using low-frequency phase as a constraint during the reconstruction process. Recent reports indicate that phase-constrained reconstruction improves image quality. The proposed method performs better than the conventional SENSE method applied slice-by-slice.