Meeting Banner
Abstract #2257

Image Reconstruction from 3D Non-Cartesian Data Employing a Combined Conjugate Gradient and Denoising Algorithm

Gregory R. Lee1, 2, Jeffrey L. Sunshine1, 2, Mark A. Griswold1, 2

1Radiology, Case Western Reserve University, Cleveland, OH, United States; 2University Hospitals Case Medical Center, Cleveland, OH, United States

Non-Cartesian 3D acquisitions, when combined with parallel imaging and compressed sensing, have great potential for accelerating MR image acquisition. However, compressed sensing reconstructions of large 3D datasets remains computationally challenging. In the present work, a simple algorithm that alternates between iterations of a conjugate gradient SENSE algorithm and a recently proposed transform-domain denoising operation is proposed. The proposed technique does not require the tuning of any regularization parameters and requires only a background noise standard deviation as input to the denoising routine. High quality reconstructions are demonstrated for contrast-enhanced MRA images undersampled by a factor of 40-150.

Keywords

able accordance acquisition acquisitions aliasing alternately alternating angiography apparent applied arterial artifacts background becomes bottom channel chose clear clearly coil combined compensation components compressed computation conjugate consent contrast convex correspond cost defaults defined density detail determined deviation dimension dimensions directly domain early either employing enhanced every fidelity filter formulates frame frames grades gradient guess hard head highly hospitals identical imaginary improvement incoherent incorporating independently initial inspired isotropic iterations iterative late listed little local makes mark medical middle minimized minor modified noise norm occurs onto parallel penalty previous previously process produced projected projection projections promoting propose proposed proved published quality radial radiology real recently reconstruct reconstruction reconstructions reduce reduction regularization regulations require required requirements requires reserve resolution respectively scanned sense sensing separately sets sharing simple solutions space sparse sparsity spatial step steps subtraction suitable sunshine superior support system table temporal term thresholding trajectories trans transform tuning typically unstable upon view volunteer western