Thanks for the info regarding these differences. I wanted to follow up with you as I looked into it and tested the conversion using dcm2nii (I believe fatcat uses this as well for importing dicoms), and saw that the magnitudes are the same for all three rows as compared with the mrtrix3 output (fsl format only). The first two rows have the opposite sign of the dcm2nii bvec output so this probably has to deal with FSL RAI (dicom order) from MRtrix3 vs LPI (neurological order) from dcm2nii. Using 1dDW_O_grad_Mat -flip_x -flip_y ended up fixing the output to match that of dcm2nii. The reason why I used mrtrix3 for importing is that we have some Philips data using the new dicom format which imported correctly using this tool and was not able to be converted using dcm2nii. I think that if I flip the signs to match dcm2nii (as well as verifying the fiber directions are in the correct orientation) this may be the fix.
Best,
Ajay