Fig. 5.

Prior distributions for the parameters of the standard model. These priors are used for generating pairs of measurements and derivatives for training the models of change. Also, the same priors are used for simulating test datasets. The priors are chosen such that they contain all probable parameter combinations that can produce measurements similar to real data. The delta function along with uniform distribution in the isotropic signal fraction is used to model pure tissue types as well as partial volume effect. In the training phase, the signal fractions are normalized to sum up to 1. A beta (shape parameters $\alpha =2,\beta =5$) distribution is used for $ODI$ to impose a nearly uniform distribution for effective fibre dispersion. The prior for isotropic and axial diffusivities are normal distributions with mean 3 and 1.7 ($\frac{\mu m^2}{ms}$) and standard deviation 0.1 and 0.3 respectively; as we expect faster diffusion as well as less variability in the free water component.