Isotropic voxels

Dear mrtrix experts,

I am using mrtrix3 and I was wondering if the algorithms automatically take into account the size of my voxels: for instance, if I have data characterized by non-isotropic voxels do the methods take this into account (this info is present in the header) or will I obtain biased results? In this case which method do you suggest to use to resize my image?

Second question: is there a limit of the b values resolution that is readable by mrtrix? Should I give to the software rounded values?

Thanks in advance for your time.


Yes, MRtrix3 will take this account.


You might, depending on how anisotropic the voxels are. See e.g. this abstract for example. But this is not something we can do anything about in the software…

I’m not sure what you’re asking…? But there should be no need to modify anything, MRtrix3 includes a few heuristics to cluster b-values together into shells in a way that should be robust to minor variations. And the values are read as floating-point values, no need to do any rounding.

Of course, if you do come across any unexpected behaviour that you think shouldn’t happen or could be done better, do let us know - if it’s a bug, we’ll typically fix it within days.

Thank you very much

I was wondering if the algorithms automatically take into account the size of my voxels

Yes, MRtrix3 will take this account.

Certainly in the context of tracking, anisotropic voxels will not lead to gross inaccuracies, since all tracking calculations are done in so-called ‘real’ / ‘scanner’ space.

There are however some e.g. image filters that are based on defining an ‘extent’ in terms of number of voxels, so in these cases anisotropic voxels will lead to anisotropic filter kernels. Something like an image gradient may be calculated ‘per voxel’ rather than ‘per mm’; this may or may not be seen as a bias, and may vary depending on exactly what type of operation you’re performing. So if there’s a specific operation outside of tracking / registration that you’re concerned about, you can try testing the result against the same image re-gridded to an isotropic voxel grid. Or if you’re really not sure, you can ask on here about that specific operation.