Meeting Banner
Abstract #3538

Multichannel Diffusion MR Image Reconstruction: How to Reduce Elevated Noise Floor and Improve Fiber Orientation Estimation

Christophe Lenglet1, Stamatios Sotiropoulos2, Steen Moeller1, Junqian Xu1, Edward J. Auerbach1, Essa Yacoub1, David Feinberg3, Kawin Setsompop4, Lawrence Walds

1Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States; 2Centre for Functional MRI of the Brain, University of Oxford, Oxford, United Kingdom; 3Advanced MRI Technologies, Sebastopol, CA, United States; 4Department of Radiology, Massachusetts General Hospital, Charlestown, MA, United States

Signal intensity in magnitude MR images follows a Rician distribution when single-channel receiver coils are employed. For multi-channel coil acquisitions, noise properties change, and the observed noise levels depend on the image reconstruction method used to combine information from different coils. This is problematic for diffusion-weighted MRI, where any artificial elevation of the noise floor limits the ability to properly quantify the signal attenuation and, ultimately, estimate fiber orientation for tractography. We propose to use a multi-channel SENSE1 reconstruction of GRAPPA un-aliased data, which exhibits Rician noise properties, and demonstrate its advantage over the RSoS reconstruction for fiber orientation estimation.

Abstract PDF

Keywords

ability achieved acquisition advanced aliasing aligned alternative although artifact artifacts attenuation ball become benefits blueprint body brain capable cause central channel channels clinical coil coils come commonly complex connect corpus correctly correlated covariance dataset degrees depend describe diffusion ding distribution educe effective either elevated elevation emerging estimation evident excitation exist expected fails fiber floor framework free freedom functional general goals gradient human illustrated impact indeed independent instead limits massive matched matrix maximal minim model modern mono motivation nine noise notice obvious ones operating orientation orientations oxford part particularly performance plotted polar polis predicting press principal profile profiles program project properties propose quantify read reasons reconstruction remain resolution retain root rove sampled sense simultaneous slice sorted spatial spurious square stick suffer support technologies terms traded unknown variance volumes whereas white