Super resolution features in MRtrix ? Merging anisotropic voxels from Cor and Tran planes

Super resolution consists of merging the DWI information from two orthogonal slices acquired with anisotropic voxel (Scherrer 2011, MICCAI).

In my case, I have acquired a 2x2x6mm3 transversal and coronal planes for 14 slices (large voxels because T2 drops in formalin fixed specimens).

Can I do this to merge the information of both acquistions :

DWI_Tran.mif and DWI_Cor.mif have first been corrected with topup feature and B1 inhomogeneities (dwibiascorrect, dwipreproc)

% flip DWI_Tran.mif (Y and Z axis) into DWI_Cor.mif coordinates
% the 4th axis is the diffusion directions
mrconvert -axes 0,2,1,3 DWI_Tran.mif DWI_Tran_flipped.mif

% merge images
% on which axis I should merge images on ?
% Mrcat does not work when .mif are not in the same dimension which can be problematic when merging anisoptropic voxels
mrcat DWI_Tran_flipped.mif DWI_Cor.mif all_DWIs.mif -axis 4

dwi2response dhollander all_DWIs.mif csd_response_tissue1.txt csd_response_tissue2.txt csd_response_tissue3.txt -voxel csd_response_voxel_investigated_tissues_1_2_3.mif

btw, thank you for making the tractography more accessible to researchers with your software !


What you’re talking about here is a slice-to-volume reconstruction. There’s no facilities to do this kind of super-resolution within MRtrix3 (yet). @bjeurissen or @dchristiaens may have their own code to do this, but it’s much more complicated than simply merging the information.

That said, you can probably merge these datasets into one if you think this is something you want to try, but you’d need to regrid each one onto the same higher resolution 2x2x2mm grid, maybe using mrresize to provide a template image at 2x2x2mm resolution (or an even smaller voxel size), and mrtransform to regrid both dataset onto the same template before mrcat – but there’s no guarantee the two datasets will be aligned, and simply merging will not super-resolve as you might hope. It’ll probably give you some resolution enhancement, since even humble linear interpolation has been shown to do that with DTI – but it’s not true super-resolution as you would expect from dedicated slice-to-volume methods.

Thank you for the hint.
I will try out and let you know