Handling FA values > 1 before voxel-wise TFCE analysis

Hello everyone,

I am fitting the diffusion tensor with MRtrix3 and have found a small number of voxels with FA values greater than 1. In every voxel where FA > 1, one of the tensor eigenvalues is negative.

The affected voxels account for approximately 0.2% of the voxels within my analysis mask, for each participant. Most appear isolated and include voxels within white matter.

My acquisition is single-shell, with one b=0 image and the remaining volumes at b=1000 s/mm². The preprocessing includes:

  • denoising with dwidenoise

  • Gibbs-ringing correction with mrdegibbs

  • susceptibility, eddy-current and motion correction using Synb0-DISCO, TOPUP and eddy_cuda --repol

  • ANTs bias-field correction

  • brain masking based on the final corrected b=0 image

The tensor and scalar maps are calculated as follows:

dwi2tensor dwi_den_preproc_unbiased.mif dt.mif \

-mask mask.mif

tensor2metric dt.mif \

-mask mask.mif \\

-fa fa.mif \\

-adc md.mif \\

-ad ad.mif \\

-rd rd.mif \\

-value eigenvalues.mif \\

-num 1,2,3

I understand that FA > 1 is possible and that the negative eigenvalue indicates a non-physical tensor estimate. However, I am unsure what the recommended practice is before conducting a group voxel-wise analysis with permutation testing and TFCE.

Thank you in advance!