Registration using mrregister

1

I need to register an unprocessed dmri image to a processed dmri image. I can’t use neither flirt nor fnirt as these are 4d-images.
Using mrregister gives the following warning:

rosella@vivi:/media/rosella/TOSHIBA EXT/dkeTest/100610/Diffusion_3T$ mrtransform unproc_3T.mif unproc_3T_transf.mif -template data_crop.nii -linear unproc2subj.mat
mrtransform: [ERROR] output file "unproc_3T_transf.mif" already exists (use -force option to force overwrite)
mrtransform: [ERROR] error creating image "unproc_3T_transf.mif"
rosella@vivi:/media/rosella/TOSHIBA EXT/dkeTest/100610/Diffusion_3T$ mrtransform unproc_3T.mif unproc_3T_transf.mif -template data_crop.nii -linear unproc2subj.mat -force
mrtransform: [WARNING] existing output files will be overwritten

and I can’t find any output.

2

If you do not provide a full path, mrtransform should create the output file relative to your current working directory. In your case, you should be able to see the file unproc_3T_transf.mif in the folder

cd "/media/rosella/TOSHIBA EXT/dkeTest/100610/Diffusion_3T"
mrview unproc_3T_transf.mif

Note that you generally do not want to transform raw DWI data with a non-rigid transformation as this is not compatible with per-volume gradient directions. For rigid transformations make sure that the gradient information is stored in the header (-grad) to be updated accordingly via the mrtransform call.

1 Like
3

I am trying to register unprocessed data to the subject dwis as I need to compute SNR in unprocessed data over specific ROIs whose parcelization is in the subject dwis space. Is this wrong?

4

If your only interest is in transforming binary mask data from one voxel grid to another, then the fact that non-rigid transformation of DWI data is not appropriate without requisite modelling is not applicable in your sepcific use case. You can still use the DWI data to drive registration if you choose - though given issues of reorientation, using something like mean b=0 images to drive the registration may be preferable - but that is distinct from whether or not it is appropriate to transform data of a particular nature once that transformation has been determined.