Meeting Banner
Abstract #2238

Combined Approximate Message Passing for Total Variation Minimization and Randomly Translated Wavelet Denoising - Improved Compressed Sensing for Diffusion Spectrum Imaging

Jonathan I. Sperl1, Ek T. Tan2, Marion I. Menzel1, Kedar Khare2, Kevin F. King3, Christopher J. Hardy2, Luca Marinelli2

1GE Global Research, Garching n. Munich, BY, Germany; 2GE Global Research, Niskayuna, NY, United States; 3GE Healthcare, Waukesha, WI, United States

Diffusion spectrum imaging (DSI) acquisition can be accelerated by randomly undersampling q-space and subsequent compressed sensing (CS) reconstruction. This work presents several extensions to CS-DSI, namely the combination of the approximate message passing (AMP) and Nesterov updates, the application of AMP to total variation minimization, and random translations for wavelet based CS. All methods can be combined yielding superior convergence properties. Fiber simulations and in vivo brain data are analyzed demonstrating the improvements in terms of speed and accuracy.

Keywords

acceleration accuracy acquisition addition allows analyzed anisotropy appears application applied applies approximate artificially authors black brain channel characterizing chosen circulant coil combination combinations combined combining compressed computation computed computes convergence conveying crossing dashed denotes depicts derivative diffusion dimension domain dotted either elements else encoding error errors fiber fibers finally fitting fraction fractional functions furthermore global going ground hardy head healthy helpful improved improvement improvements incorporated interval invariance invariant iteration iterations iterative kurtosis like long maps math message metrics minimization mixture model modifications moreover noise noiseless noisy operator optimally orientation original orthogonal outside passing patterns phys prof properties proposed pure random randomly recent reciprocal reconstructed reconstruction reconstructions reduces replaced resolving respectively sampled sampling scanner sensing several shrinkage simplified simulation simulations slightly soft solid solved space spectrum still structures subsequently substantially superior tensors term thank theory thresholding transform translated translations truth updates variation various visible volunteers wavelet wavelets whereas worst yielded yielding yields