Hi, I would leave the values as they are for the comparison, as the estimates will have less bias this way. All the best, Mark On 20 Feb 2014, at 16:38, charlie Leger <[log in to unmask]> wrote: > There seem to be several thread explaining negative MD values and FA>1, and it makes sense that such values can result from noise and perturbed eigenvalues of the reconstructed tensor. > > However, in say, a comparison of DTIFIT FA and MD values between two time points (same patients) should the negative MD values and FA >1 be converted with scalar values, scalar values that reflect the contribution of noise?