Registering and warping FOD map to another study FOD template

Dear MRtrixperts,

I have data from 2 studies: one single-shell, and the other multi-shell. I am trying to register and warping the FOD map from one study to the other study’s template to visually compare differences between these studies and to appreciate whether single-shell acquisitions are acceptable relative to multi-shell acquisitions. To do so, I have run mrregister with the wmfod from the single-shell study, the mask1 option, and the template from the multi-shell study. I also outputted the non-linear deformation fields with nl_warp. Then, I applied those transformation to create the FOD map in template space using mrtransform.
The mrregister step computed without error, but there is obviously a problem with my FODmap in template space overlaid on the wmfod_template. I think the problem comes from the mrregister step. I was wondering if it’s a problem that the 2 studies do not have the same voxel sizes? Plus, the single-shell study has anisotropic voxel size.

thank you for your help!


That should not be a problem as long as there is sufficient spatial information in all dimensions. Internally mrregister operates on a voxel grid that lies between that of both images in terms of orientation and voxel size.

Your ODF overlay looks somewhat fishy: It appears that multiple ODFs are overlaid onto each other and the voxel size and grid orientation of the transformed and target (assuming you used it for the -template option in mrtransform) should match that of the target. Does the first volume of the transformed image look alright when opened as an image in mrview? Does the non-transformed image look alright as ODFs when overlaid onto itself?

Without knowing the exact commands you used, it is hard to say whether it is a problem with the registration result or with the transformation command. To test this, the easiest would be to repeat the registration with source and target image as 1st and 2nd arguments and using the -transformed option, which should give you image 1 transformed into the space of image 2. Does that image contain the full brain and if so, does it overlay with the target image (overlay tool or flick between images in mrview)? If not, then registration went wrong. Then, check that intensities in both images are comparable, check the masks if you used any and the option -type rigid_affine_nonlinear might help.