Bayesian estimation applied to stretched exponential modelling of diffusion-weighted imaging data

The stretched exponential model provides a means of assessing non-monoexponential signal attenuation in diffusion-weighted imaging data. However, standard least squares approaches to parameter estimation remain susceptible to noise. In this work, we apply two distinct Bayesian approaches to stretched exponential modelling, and assess their performance at low-medium b-values using simulations. The use of a spatial homogeneity prior is found to yield parameter maps with higher precision and accuracy than those obtained with the use of a Gaussian shrinkage prior or nonlinear least squares.