Bayesian methods accurately predict ADC bias resulting from clinical diffusion-encoding gradients: Validation through simulation studies.

Motivation: Measurement of ADC in body DWI is typically assumed to be isotropic and therefore the use of single-direction diffusion encoding imaging is commonplace. Estimation of induced bias by this assumption is needed. Goal(s): To determine whether Bayesian estimation of ADC measurement bias from DTI data suffers from any systematic errors and thus can be used reliably in clinical datasets. Approach: We use simulation studies over a range of true fractional-anisotopy (FA), signal-to-noise ratio (SNR) and mean-diffusivity parameters, and investigate the accuracy of Bayesian estimation approaches. Results: Bayesian estimation of ADC bias appears accurate over the range of tested parameters. Impact: Accurate estimation of ADC bias from single-direction diffusion-encoding schemes is possible using Bayesian approaches in combination with data acquired using a multi-directional diffusion-encoding scheme. This enables pixel-wise estimation of bias and could negate the need for directly acquiring paired datasets.