Diffusional kurtosis imaging (DKI) extends conventional diffusion tensor imaging (DTI) by estimating the kurtosis of the water diffusion probability distribution function1-4. The kurtosis is a general, dimensionless statistic for quantifying the non-Gaussianity of any distribution5. A positive kurtosis means the distribution is more strongly peaked and has heavier tails than a Gaussian distribution with the same variance. Water diffusion in biological tissues is non-Gaussian due to the effects of cellular microstructure (e.g., cell membranes and organelles). This is particularly evident in brain, where water diffusion is strongly restricted by myelinated axons. Qualitatively, a large diffusional kurtosis suggests a high degree of diffusional heterogeneity and microstructural complexity.
Because diffusion in brain is anisotropic, DKI requires the introduction of a diffusional kurtosis tensor in addition to the diffusion tensor used in DTI. From the diffusion and diffusional kurtosis tensors (which are calculated together from a single diffusion-weighted imaging dataset), several rotationally invariant metrics can be computed. These include standard DTI metrics, such as the mean diffusivity and fractional anisotropy, as well as metrics reflecting the diffusional kurtosis, such as the mean, axial, and radial kurtoses. The diffusional kurtosis metrics are strongly linked to cellular microstructure, as this is the main source of diffusional non-Gaussianity in tissues. The extra information provided by DKI can also resolve intra-voxel fiber crossings and thus be used to improve fiber tractography of white matter 6,7.
An advantage of DKI is that it is relatively simple to implement for human imaging on conventional MRI clinical scanners. DKI protocols differ from DTI protocols in requiring at least 3 b-values (as compared to 2 b-values for DTI) and at least 15 independent diffusion gradient directions (as compared to 6 for DTI). Typical protocols for brain have b-values of 0, 1000, 2000 s/mm2 with 30 diffusion directions. Image post-processing requires the use of specialized algorithms4,8.
DKI has been most commonly used for the study of the brain9-52, although applications to other body parts such as lung53, head and neck tumors54, prostate cancer55-58, breast cancer59, calf muscle60 and the liver61 have been investigated. There are also several technical studies involving DKI62-71.