Richard A E Edden1,2, Derek K. Jones3
1Russell H Morgan Department of Radiology and Radiological Science, The Johns Hopkins University, Baltimore, MD, United States; 2FM Kirby Research Center for Functional MRI, Kennedy Krieger Institute, Baltimore, MD, United States; 3CUBRIC, School of Psychology,, Cardiff University, Cardiff, Wales, United Kingdom
DTI provides rotationally invariant information. Additionally, DTI acquisitions are optimised to ensure that data are statistically rotationally invariant so that parameter variance is independent of the orientation of the fibre population within the brain. Against this backdrop, we focus on skeletonization-based methods for group comparisons of DTI data and show that they can reintroduce rotational dependence. Specifically, the power to detect group differences in a fibre can depend on its orientation. While the cause/solution to this problem are trivial, the effect on statistical inference is not and should be viewed in the light of the increasing popularity of skeletonization-based methods.