PuSH - Publikationsserver des Helmholtz Zentrums München

Hahn, K.R. ; Prigarin, S.M.* ; Rodenacker, K. ; Hasan, K.M.*

Denoising for diffusion tensor imaging with low signal to noise ratios: Method and Monte Carlo validation.

Int. J. Biomath. Biostat. 1, 63-81 (2010)
Verlagsversion bestellen
Diffusion Tensor Imaging (DTI) is a magnetic resonance technique which enables the in vivo visualisation and characterisation of nerve fibers in the human brain indicating diseases like Multiple Sclerosis or Alzheimer. Increasing the sensitivity by a reduction of the voxel size or by application of measurement parameters with better diffusion differentiation induces low signal to noise ratios (SNR) in the measured diffusion weighted images (DWI) which collect the information for DTI. Due to nonlinear noise propagation within the DTI formalism the fiber properties are modelled by biased non Gaussian random fields. We present a new denoising method for DTI with low SNR and explore it by Monte Carlo simulations based on a phantom and on human brain data with high resolution 1 mm3 voxels. The method combines voxelwise averaging, nonlinear spatial filtering and a modified Rician bias correction of the DWIs to reduce bias and noise in the estimated images of fiber properties. These procedures reduce the main part of the mean squared error, but due to sample size restrictions residual noise may remain. Therefore, (post)filtering is finally applied directly to those denoised images. The method is linked to the Delta Method formalizing the asymptotic Gaussian limit of nonlinear noise propagation. The simulations demonstrate the feasibility of a quantitative analysis of DTI data of the human brain with 1 mm3 resolution, measured at a clinical field strength of 3T, if thermal noise is the source of the dominating artifacts. In addition, the results support the design of new high resolution experiments. The method is not limited to the standard tensor model.
Weitere Metriken?
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Journalartikel
Dokumenttyp Wissenschaftlicher Artikel
ISSN (print) / ISBN 0973-7340
Quellenangaben Band: 1, Heft: 1, Seiten: 63-81 Artikelnummer: , Supplement: ,
Verlag Serials Publ.
Begutachtungsstatus Peer reviewed