Meeting Banner
Abstract #3590

Fast DSI Acquisition and Reconstruction Based on Sparse Diffusion Propagator Representations

Antonio Tristn-Vega1, Carl-Fredrik Westin1

1Laboratory of Mathematics in Imaging, Brigham and Women's Hospital, Boston, MA, United States

Compressed Sensing grants the possibility of reducing the number of samples required to describe a signal below its Nyquist rate. When applied to the estimation of the diffusion propagator in diffusion spectrum imaging, it arises the need to find a basis for which the propagator is sparse, which is not trivial. We address this problem by two means: 1) mapping the propagator to a space where it is sparse, and 2) using a model that explicitly isolates a non-sparse residual. We provide two reconstruction algorithms for such model, notably improving the estimation accuracy and sparsity compared to previous approaches.

Keywords

able accelerated accounts acquisition address approximation assuming basis behavior canonical china claim compact compressed computed conclude considerably considerations considered consistent contrary converges crossing detect diffusion direct discrete discretized eigenvalues either energy enforcing ensemble equal equivalent error errors estimating examples expected expression faithful fibers fixed frame frequency fulfill global greatly hard hence highly illustrates improving included inclusion inverse kept laboratory larger limitation local mathematics matrix minimization mixture model nearly need noise note novel operations opposed optimization outliers overall part partial previous problem producing propagator properly proposing proven pseudo quadratic randomly realistic reconstructed reconstruction reconstructions reduces reducing related replicas represent representations representative represented required residual restoration rotated sampled samples sampling scenario seems selects sensing simulate simulated since situation smooth soft solution solved sparse sparsely sparsest sparsity spectrum step steps strictly suboptimal suitable table term theory though thresholding tight together unless unpractical usually vector wave wavelet wavelets whenever whose wide women workshop worth zeros