Leo J. Grady1, Jonathan R. Polimeni2
1Imaging and Visualization, Siemens Corporate Research, Princeton, NJ, USA; 2NMR Center, Radiology Department, Massachusetts General Hospital, Charlestown, MA, USA
Compressed sensing techniques have recently become very popular for image reconstruction given sparsely sampled data. However, previous methods only approximately enforce the constraint that the reconstructed image has Fourier coefficients at the sampled locations, which require manual parameter tuning and result in slower speeds for convergence. We present a fast, parallelizable method that it is capable of quickly reconstructing a sparsely sampled image that exactly satisfies the acquisition. We demonstrate that our technique is able to produce an exact reconstruction of the Shepp-Logan phantom with very sparsely sampled k-space data for realizable accelerated acquisitions using two common sampling trajectories.