Advanced fit of the diffusion kurtosis tensor by directional weighting and regularization

Abstract

The diffusional kurtosis is an indicator for diffusion restrictions in biological tissue. It is observed experimentally that the kurtosis is largest for directions perpendicular to the fiber direction in white matter. The directional dependence of the kurtosis can be described by the diffusion kurtosis tensor. Since the intention of diffusion kurtosis imaging is to detect diffusion restrictions, the fit of the kurtosis tensor should be dominated by directions perpendicular to the fibers. In this work, it is shown that the basic approach, which is solving the occurring linear system by a pseudoinverse matrix, may completely fail in this regard if the diffusion is highly anisotropic. This problem is solved by adapting the weights of the fit--and thus emphasizing directions of restricted water motion--using a direct fit of the kurtosis tensor to the measured kurtosis values. Moreover, due to its large number of degrees of freedom, the kurtosis tensor can assume complicated shapes resulting in a fit which is sensitive to noise. This article demonstrates that the quality of the kurtosis tensor calculation can be further improved if the fit is regularized by suppressing too large and too small kurtosis tensor values and thus restricting the possible tensor shapes.