Meeting Banner
Abstract #3580

DWI Denoising Using Overcomplete Local PCA Decomposition

Jose V. Manjon1, Pierrick Coupe2, Luis Concha3, Antoni Buades4, Louis Collins5, Montserrat Robles6

1IBIME, UPV, valencia, Spain; 2LaBRI, Bourdeaux, France; 3UNAM, Mexico; 4Universite Paris Descartes, France; 5MNI, Canada; 6IBIME, UPV, Valencia, Spain

Diffusion Weighted Images normally show a low SNR due to the presence of noise from the measurement process which complicates and potentially bias the estimation of the diffusion parameters. In this paper, a new denosing method is proposed which takes into consideration the multicomponent nature of DW images. This new filter reduces random noise in multicomponent Diffusion MR images by locally shrinking less significant Principal Components using an overcomplete approach. The proposed method is compared with similar state of art methods using synthetic and real clinical MR images showing an improved performance in all cases analyzed.

Abstract PDF E-Poster Video

Keywords

acquisitions advantage allows anal applied around artifacts averaging better brief channel column common components computed consistently corresponds coupe decomposition designed diffusion elements example filtered filtering finally hand impact improve improved inherent intensity joint local matrix minimizes natural noise noisy original overlapping particular patch patches pattern phantom pixel prior processed propose proposed quality real recompose reconstruction reducing redundancy removing reordered residuals several significantly simulations soft step subtracting surrounding synthetic take tensor thermal thresholding upper variance variants vector windows