Independent component analysis applied to diffusion tensor MRI

Abstract

The accuracy of the outcome in a diffusion tensor imaging (DTI) experiment depends on the acquisition scheme as well as the postprocessing methods used. In the present study, the DTI results acquired after applying different combinations of diffusion-weighted (DW) gradient orientations were initially compared. Then, spatially independent component analysis (ICA) was applied to the T2 and DW images. In all cases a single component was detected that was similar to the map of the trace of the diffusion tensor, but contained a reduced amount of noise. Furthermore, when no correction for eddy current artifacts was used in the image acquisition scheme, the effects of eddy currents were separated by ICA into independent components. After these components were removed, conventional estimation of the diffusion tensor was performed on the modified data. No artifact was contained in the final rotationally invariant scalar quantities that describe the intrinsic diffusion properties. Additionally, independent components that mapped major white matter fiber tracts in the human brain were identified. Finally, the noise included in the original T2 and DW images was also separated by ICA into independent components. These components were subsequently removed and a reduction of noise in the final DTI results was achieved. © 2002 Wiley-Liss, Inc.