In thinking about appropriate definitions of orientation coordinates for DWI derivatives, I realised that in its current form the main BIDS spec makes the same mistake that we made in MRtrix3 regarding the definition of the FSL bvecs / bvals format.

The best explanation of the issue was posted on our community forum; the FSL FDT documentation has also since been updated accordingly.

Long story short

The current description is:

Inherent to the FSL format for bvec specification is the fact that the coordinate system of the bvecs is with respect to the participant (i.e., defined by the axes of the corresponding dwi.nii file) and not the magnet’s coordinate system

This is not precisely correct. The vector orientations defined in the bvecs file are with respect to the i,j,k axes of the corresponding NIfTI image as though the NIfTI were to use a neurological convention. That is: regardless of whether the associated NIfTI image is LAS (Analyze heritage; negative transform determinant) or RAS (NIfTI spec), the bvecs directions are with reference to the LAS equivalent of that NIfTI image.