Meeting Banner
Abstract #3219

A Simple and Efficient Method for Acceleration and Denoising of Multi-Contrast Diffusion Data: Application to Q-Space and HARDI

Ana-Maria Oros-Peusquens1, Nadim Jon Shah1, 2

1INM-4, Research Centre Jlich, Jlich, Germany; 2Faculty of Medicine, JARA, RWTH Aachen University, Aachen, Germany

We present a method combining acceleration of multi-contrast diffusion data using UNFOLD with denoising based on singular value decomposition (SVD). UNFOLD is ideally suited for acceleration of q-space acquisition because q-space information is extracted from the Fourier transform of the q-dependent diffusion signal. The information is largely contained in the q-space centre, a region unaffected by aliasing at moderate UNFOLD acceleration (2-4). Multi-direction HARDI signal also has Fourier components in a restricted interval and is thus very well suited for UNFOLD acceleration. SVD denoising works extremely well for q-space data and also, but less spectacularly, for quantitative information obtained with HARDI. In conclusion, acceleration and denoising of q-space and HARDI data is demonstrated, using simple methods with numerically undemanding reconstruction. The same method should work well for combinations of q-space and HARDI, such as diffusion spectrum imaging (DSI).

Abstract PDF Presentation Video

Keywords

accelerated acceleration accessible account acquisition already amplitudes appearance application applications appreciated approved array attenuation axis beneficial better body brain capable central clearly coil committee component components concentrated conjunction consuming contrast contrasts correspond crucial decay decomposition detection deviation diffusion diffusivity dimension dimensional displacement dissected distribution doubly dramatic efficient eliminated equipped ethics even every extent extract extracted extremely fast fitting fixed full fully glover good gradient half ideally identical illustrate in vivo included indeed individual influence informed insights inverse investigate involved keeping kurtosis largest last linear local maps maria matched matrix medicine model moderate modified narrow necessary need noisy original piece plotted polynomial potential practically preparation probability procedures promise propagator receiver reconstructed reconstruction reduced relevant represent resolution resorting respectively retrieved roughly sample sampled samples scale scanner securing shah shape shortcomings simple singular slice slow space spaces spectra spectral spectrum stack starting structural terms tissue transform transmit trio truncation unaffected unfold unfolded visual volunteer zero zeroed