Spatial Smoothing for Diffusion Tensor Imaging with low Signal to Noise Ratios

Though low signal to noise ratio (SNR) experiments in DTI give key
information about tracking and anisotropy, e.g. by measurements
with very small voxel sizes, due to the complicated impact of
thermal noise such experiments are up to now seldom analysed. In
this paper Monte Carlo simulations are presented which investigate
the random fields of noise for different DTI variables in low SNR
situations. Based on this study a strategy for spatial smoothing,
which demands essentially uniform noise, is derived. To construct a
convenient filter the weights of the nonlinear Aurich chain are
adapted to DTI. This edge preserving three dimensional filter is
then validated in different variants via a quasi realistic model
and is applied to very new data with isotropic voxels of the size
1x1x1 mm3 which correspond to a spatial mean SNR of approximately
3.