Meeting Banner
Abstract #2268

Automated RF Spike Noise Removal with Compressed Sensing

David S. Smith1, Ryan Robison, 12, E Brian Welch1

1Institute of Imaging Science, Vanderbilt University, Nashville, TN, United States; 2Philips Healthcare, Cleveland, OH, United States

We present a method using concepts from compressed sensing to automatically detect and eliminate radio frequency spike noise from MRI data sets. The spikes are located by their effect on the total variation of the image. The spikes are then deleted from the full data set, creating a very slightly undersampled data set, which is then reconstructed in a TV-regularized compressed sensing MRI reconstruction. Since the data is almost completely Nyquist sampled, this method introduces no artifacts and produces images with normalized mean square error two orders of magnitude smaller than both zeroing of the spiked data and Fourier interpolation.

Keywords

accurate added advantage amplitude anisotropic approaches artifacts automated automatic automatically becomes beside bore bottom bursts cable cause channel clean cleaned clear compensation complexity compressed compressive constrained corrected correction correctly corrupted corruptions create created creating criterion defined detected detection discarded discrete domain dynamic element eliminated equal error example experiment faulty five form fraction free frequency full gains generated gradient ground identity include inconsistencies inserted institute interpolated interpolating interpolation inverse largest lead leading leftmost legs limit linearly located locates locations magnitude mask materials matrix middle mostly must noise noiseless normalized original outperform outperforms partial pattern patterns phantom proposed quality radio reasonable reconstructed reconstruction reconstructions regularized reliability remaining removal removed replaced replacement replacing residual samples sampling scalar scanner science searched sensing setting significantly simulated slightly smallest smith sorted space spatial spike spikes square stimulating superiority thank thresholding traditional transform truth turn unconstrained unknown upper variable variation weighed zero zeroed zeroing zeros