Affordable Access

SNR dependence of mean kurtosis and how to correct it

Publication Date
  • Human Health Sciences :: Laboratory Medicine & Medical Technology [D13]
  • Sciences De La Santé Humaine :: Médecine De Laboratoire & Technologie Médicale [D13]
  • Medicine


untitled 737 SNR dependence of mean kurtosis and how to correct it Elodie André1, Christophe Phillips1,2, Ezequiel Farrher3, Ivan Maximov3, Farida Grinberg3, N. Jon Shah3,4, and Evelyne Balteau1 1Cyclotron research center, University of Liège, Liège, Belgium, 2Department of Electrical Engineering and Computer Science, University of Liège, Liège, Belgium, 3Institute of Neuroscience and Medicine - 4, Forschungszentrum Juelich GmbH, Juelich, Germany, 4Department of Neurology, Faculty of Medicine, RWTH Aachen University, JARA, Aachen, Germany Target audience This abstract evaluates procedures for noise reduction and noise bias correction in MR images and their impact on DKI estimates. It is of interest to scientists processing diffusion-weighted and other low SNR MR images. Introduction In diffusion imaging, low signal-to-noise ratio (SNR) becomes critical, especially at high diffusion weightings (b-values). Complex images resulting from the combination of multichannel data are generally rendered as magnitude images, introducing a strong bias in the actual signal estimate at low SNR. This is particularly crucial for diffusion kurtosis imaging (DKI). Two correction methods are presented here: (1) power image correction1 adapted for multichannel data and (2) look-up table correction based on the analytical expression of the central chi distribution of the noise2. The dependence of mean kurtosis (MK) on SNR is discussed in a first experiment while the inter-subject MK variability is discussed in a second experiment. Materials and methods Diffusion-weighted (DW) data were acquired on a 3T scanner (Allegra, Siemens Medical Solutions, Erlangen Germany) with an 8-channel receive head coil using a twice-refocused-spin-echo diffusion sequence. To study the noise correction and SNR impact on DKI results, two experiments were carried out. Protocol 1 (Number of repetitions (NR)=1, b=0/1000/2500 s/mm2; TR/TE=7400/91 ms, FoV= 211 mm, matrix 88x88, voxel size 2.4x2.4x2.4 mm3) was repea

There are no comments yet on this publication. Be the first to share your thoughts.


Seen <100 times