Effect of mixed fibre fraction and orientation samples near crossing-fibre areas

Fibre fraction parameters (f₁, f₂, etc.) and their associated orientations (ϕ₁θ₁,ϕ₂θ_2, etc.) could be inconsistently associated with the different underlying sub-fibre populations, especially if the fibre fractions are of comparable strength [27]. This can cause differing proportions of fibre fraction and orientation values to be labeled as one group (e.g. f₁,ϕ₁θ₁) but labeled as another on the next trial (e.g. f₂,ϕ₂θ₁). There is no guarantee that the labeling happens consistently and because we are merging samples from 20 different bedpostx and bedpostx_gpu trials to form the PDF distributions for comparison, it is possible that differences between the two platforms occur due to the this inconsistent labeling of sub-fibre populations. To investigate this effect of mixed fibre fractions and how much it may contribute to CPU and GPU output differences, we swapped f₁,ϕ₁θ₁ and f₂,ϕ₂θ₂ where f₂ > f₁ and ran the same statistical analysis on the swapped samples and compared the results against statistically different unswapped samples.