FSL fits a linear model to the log of the signal to obtain the tensor. It does not impose any positive definiteness constraints on the tensor. Therefore, negative eigenvalues can be obtained in the presence of noise, leading to a larger than 1 FA value.
On 27 Apr 2010, at 14:15, Marius Widerĝe wrote:
> I have used DTIFIT to calculate FA-maps from diffusion data of a rat brain.
> On the border between the brain tissue and surrounding tissue (CSF, meninges etc) FA values higher than 1 is reported. How is this possible?
> Example of one voxel: Reported FA-value: 1.22429. Corresponding L1=0.00033, L2=0.000018 and L3=0.00029. When I calculate the FA myself in this voxel based on the reported L1,L2 and L3 I get 0.668767.
> I am aware that these voxels are probably noise and should be disregarded, but it worries me that FSL actually reports FA-values higher than 1.
> Can anyone explain what has happened here?