# Data Analysis

## Data Processing

The acquired images are registered using FMRIB Software Library (FSL, FMRIB, Oxford) rigid-body registration [24] to account for any potential vibrations from the oscillating gradient waveforms. The SNR is calculated from the mean and standard deviation across 9 b = 0 images per voxel. The region of interest (ROI) is chosen from the b = 0 images by manually excluding edges of the plate to avoid voxels affected by partial volume effect. The ROIs of all plates has a mean SNR > 45. Additionally, as the direction of the microcapillaries, n, is assumed to be unknown, n is estimated using OGSE diffusion tensor imaging and then is inputted into our model fitting procedure described below.

## ActiveAx Model Fitting

A voxel-wise two stage model fitting procedure, as defined in [11], is used to estimate diameter and diffusivity of the plate samples. The procedure consists of an initial grid search followed by a non-linear fitting of the model described in (1)to the measured signal. The gradient descent optimisation is constrained with user-defined limits for all parameters (lower and upper bound limits: *a* = 0.002 and 30.0 |xm, and *Dj* = 0.002 and 6.0 |i,m^{2}/ms, respectively) to speed up the fitting. The microcapillary diameters and diffusivities are estimated for each voxel in the ROI, then the mean and standard deviation of the parameters for each plate are calculated across the chosen ROI.