Monotonic non-linear transformations as a tool to investigate age-related effects on brain white matter integrity: A Box–Cox investigation

Abstract

Non-linear effects of age on white matter integrity are ubiquitous in the brain and indicate that these effects are more pronounced in certain brain regions at specific ages. Box-Cox analysis is a technique to increase the log-likelihood of linear relationships between variables by means of monotonic non-linear transformations. Here we employ Box-Cox transformations to flexibly and parsimoniously determine the degree of non-linearity of age-related effects on white matter integrity by means of model comparisons using a voxel-wise approach. Analysis of white matter integrity in a sample of adults between 20 and 89years of age (n=88) revealed that considerable portions of the white matter in the corpus callosum, cerebellum, pallidum, brainstem, superior occipito-frontal fascicle and optic radiation show non-linear effects of age. Global analyses revealed an increase in the average non-linearity from fractional anisotropy to radial diffusivity, axial diffusivity, and mean diffusivity. These results suggest that Box-Cox transformations are a useful and flexible tool to investigate more complex non-linear effects of age on white matter integrity and extend the functionality of the Box-Cox analysis in neuroimaging.

Age is reflected in the Fractal Dimensionality of MRI Diffusion Based Tractography

“…5 ). Since the existence of non-linear age-related effects is well documented 21 , 37 , 38 , linear regression and additional polynomial regression was employed to describe the effect of age on FA and FD. In both cases, R 2 was higher for FD than for FA (linear fit FD : R 2 = 0.520, linear fit FA: R 2 = 0.214; polynomial fit FD : R 2 = 0.576, polynomial fit FA: R 2 = 0.210).…”

An attempt to model the causal structure behind white matter aging and cognitive decline

“…This study is a reanalysis of a previous published DTI study 27 . All the descriptions of the sample composition, sample properties and scanner parameters of the study mentioned above apply to this study as well.…”

Fat tails and the need to disclose distribution parameters of qEEG databases

“…Therefore, one should ideally always determine the transformation capable of adequately bending the shape of the distribution of qEEG data into the Gaussian. One practical way to do that is to use other families of transformations [10,32,33]. However, this step of data processing is not trivial, since estimating distribution moments and properties when distributions are more pronouncedly fat-tailed requires much more data than the usual few hundreds or thousands available in qEEG data bases.…”