Meeting Banner
Abstract #2077

Low-Rank Basis Smoothing for the Denoising of Diffusion Weighted Images

Stephen F. Cauley1, Obaidah A. Abuhashem2, Berkin Bilgic3, Itthi Chatnuntawech3, Julien Cohen-Adad4, Kawin Setsompop5, Elfar Adalsteinsson2, 5, Lawrence L. Wald5, 6

1A.A. Martinos Center for Biomedical Imaging, Dept. of Radiology, Massachusetts General Hospital, Charlestown, MA, United States; 2Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA, United States; 3Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States; 4Department of Electrical Engineering, Institute of Biomedical Engineering, Ecole Polytechnique de Montreal, Montreal, QC, Canada; 5A.A. Martinos Center for Biomedical Imaging, Dept. of Radiology, MGH, Charlestown, MA, United States; 6Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, United States

The substantial signal attenuation in DI images for large b-values can affect accurate calculation of orientation distribution functions (odf) and fiber tracks. In addition, the low signal-to-noise (SNR) observed at large b-values hinders the performance of popular denoising methods like the LMMSE estimator and the NLM filter. In this work we demonstrate the benefits of basis smoothing within a low-rank DWI estimation framework. Our method significantly reduces the dependencies on noisy basis vectors while preserving root-mean-square error (RMSE) relative to low-noise data (computed by averaging multiple acquisitions).

Abstract PDF

Keywords

able accurate acquisitions addition affect analyze approaches arranged attenuation audience available averaging basis benefits biomedical calculation central column computationally computed computer computing consisting constraints contained contains correlation correlations correspond corrupting critical damping decay decomposition dependencies dependency dept diffusion distribution division dominant edge effectively efficient electrical employed engineering error estimation estimator exploited fact features fiber filter form framework functions health healthy hinders implementations important imposing improving institute introduced investigators isotropic joint like listed local locations maintain many matrix minimize name noise noisy note observations onto orientation performance popular preserving produce project projected projection publicly quality radiology rank reduced reduces reducing reduction reflected regularized relatively resolution retains root scanner science sciences several significantly simplified singular smallest smoothed smoothing somewhat space speckling square still strategy structure subsequently substantial substantially support target technology threshold throughout toolbox tracks truncating tuned unable unlike vector vectors version versions volunteer