Diffusion encoding

1

Does diffusion encoding information of dmri images change after they are nonlinearly normalized to MNI space?
thank you
Rosella

2

Yes, it would – and in a spatially-dependent manner. I would typically recommend doing all the processing in subject space, and only registering the results to template space. We have tools to do that, but you need to be careful with registration of diffusion MRI data.

1 Like
3

What about denoising? Does a denoising procedure change diffusion information?
thank you

4

No it doesn’t. Denoising does not change the spatial localisation and is completely independent of the gradient encoding.

1 Like
5

To clarify:

Does diffusion encoding information of dmri images change after they are nonlinearly normalized to MNI space?

Yes, it would – and in a spatially-dependent manner.

  1. It is not that this is what would happen if you were to do this in MRtrix3: this is something that would ideally need to happen as the result of such a process, but is not currently supported.

  2. Non-linear warping of diffusion image information would not only alter the gradient orientations in a spatially-dependent manner. It also ideally requires some form of modulation, to reflect the fact that non-linear transformation can spatially expand or contract particular structures. This shouldn’t be done on the DWI signal directly, and instead requires some form of model; see for example the GQSI model. In MRtrix3 land this was instead addressed originally by first applying the CSD model and then modulating the FODs during transformation; nowadays we instead compute the FC metric during the FBA pipeline, which directly quantifies how much non-linear spatial transformation affects the capacity for fibre structures to transfer information. But both of these apply to fibre orientation information; the same mathematics do not apply to diffusion-weighted image information.