I was wondering if MRtrix3 contains a specific spatial smoothing filter during the preprocessing steps, which is known as the ‘anisotropic diffusion filter’ (ADF)? It is based on the Perona-Malik algorithm, and it is known to be an alternative with less blurring compared to the Gaussian kernel for spatial smoothing. I know that both mrfilter and fslmaths has the Gaussian smoothing method option, but I haven’t seen the ADF method in either of them. I know that AFNI has this option, and I have seen other source codes of it online, but was wondering if MRtrix has it as well.
No, I’m afraid we haven’t implemented that – sorry.
I was wondering if MRtrix3 contains a specific spatial smoothing filter during the preprocessing steps …
Worth interrogating briefly the rationale behind the request:
As far as smoothing prior to statistical inference is concerned, the combination of a spatial kernel with fixel connectivity information means that “the smoothing will be anisotropic”.
For anisotropic incorporation of neighbourhood information for trying to produce cleaner FODs than performing that operation on a purely voxel-wise basis, @jdtournier did create the “robust CSD” method some time ago. But it’s never been released as part of MRtrix3, for reasons that are far too esoteric to go into here…
No idea if either of these are relevant, just thought I’d include in case the specificity of the question was a red herring.
Of course for anyone interested in learning the MRtrix3 API, implementation of established methods like this are a great starting point