Indices derived from diffusion tensor imaging (DTI) data, including the mean diffusivity (MD) and fractional anisotropy (FA), are often used to better understand the microstructure of the brain. DTI, however, is susceptible to imaging artefacts, which can bias these indices. The most important sources of artefacts in DTI include eddy currents, nonuniformity and mis-calibration of gradients. We modelled these and other artefacts using a local perturbation field (LPF) approach. LPFs during the diffusion-weighting period describe the local mismatches between the effective and the expected diffusion gradients resulting in a spatially varying error in the diffusion weighting B matrix and diffusion tensor estimation. We introduced a model that makes use of phantom measurements to provide a robust estimation of the LPF in DTI without requiring any scanner-hardware-specific information or special MRI sequences. We derived an approximation of the perturbed diffusion tensor in the isotropic-diffusion limit that can be used to identify regions in any DTI index map that are affected by LPFs. Using these models, we simulated and measured LPFs and characterised their effect on human DTI for three different clinical scanners. The small FA values found in grey matter were biased towards greater anisotropy leading to lower grey-to-white matter contrast (up to 10%). Differences in head position due to e.g. repositioning produced errors of up to 10% in the MD, reducing comparability in multi-centre or longitudinal studies. We demonstrate the importance of the proposed correction by showing improved consistency across scanners, different head positions and an increased FA contrast between grey and white matter.
