When using dwifslpreproc with a simple pair + dwi combo, I noticed that applytopup after topup is not correctly applying the field. The distortions are made worse instead of less and the resulting brain mask suffers as a result from that.

Looking into this, I noticed that mrinfo -export_pe_table produces a txt file that has the phase encoding direction flipped as opposed to what mrinfo -export_pe_eddy produces. Since applytopup is using the txt file from mrinfo -export_pe_table, it appears to apply the opposite correction. Subsequent eddy on the other hand is using the txt file from mrinfo -export_pe_eddy, so it seems to apply the field correctly. (NOTE: the issue turned out not to be pe_table vs pe_eddy, but mrinfo vs mrconvert with strides).

With dwi.mif containing a single PE direction:

mrinfo dwi.mif -export_pe_table pe_table.txt -export_pe_eddy pe_eddy.txt pe_eddy_idx.txt -force
cat pe_eddy.txt

produces

0 1 0 0.0321

whereas

head -n 1 pe_table.txt

produces

 0 -1  0 0.0321

The cause seems to be that -export_pe_eddy boils down to save(transform_for_nifti_write(PE, header), path), which includes