Analyzing WM integrity with MRtriX


Hello Mrtrix Team,
for a study I would like to compare DWI data of two groups, one is doing sport and the other not and only serve as comparison. I would like to analyse the WM integrity. I found on your webpage different topics and terms, such as AFD, FC, and FDC. However I have still some questions:

  1. What would you suggest to do for such project? Doing the fixel based analysis as described in:
    or is it enough to calculate and compare AFD maps via fod2fixel and fixel2voxel, or is there another approach implemented in MRtriX?
    2.What are exactly the differences between the metrics AFD and FC, FC and FDC?
  2. The fixel based tutorial includes also tractography (19-22), are these steps necessary in case of a WM integrity comparison?

Thank you for your support!!!


Hi Max,

  1. In the absence of an a priori hypothesis, a whole-brain fixel-based analysis (FBA) is generally the recommended approach. Using fod2fixel then fixel2voxel would result in (most likely) spatial maps of what we refer to as “AFD_total”, or the total (apparent) fibre density in each voxel. This process collapses potentially informative crossing-fibre information into a voxel-wise measure; Furthermore. some approach would still be required in order to perform statistical inference; while “conventional” voxel-cluster-based statistical inference could be used, there are particular benefits to the Connectivity-based Fixel Enhancement (CFE) method that is used as part of the FBA pipeline when crossing-fibre information is preserved.

  2. “Apparent Fibre Density (AFD)” was the original term used to describe this quantitative measure in its first demonstration. However with the further developments that became what we now refer to as FBA, the terminology required refinement, which lead to the metrics FD, FC and FDC, as described in the FBA paper. While the term “AFD” can still be sometimes used, typically to refer more generally to the use of quantitative metrics derived from CSD, whenever specificity of FBA metrics is required the terms FD / FC / FDC must be used.

  3. The use of tractography in the FBA pipeline is purely for deriving a tractogram in population template space for the purpose of statistical enhancement: this information is integral to the CFE method. However tractography is not applied at the per subject level in this pipeline.
    Moreover, I would strongly discourage use of the term “WM integrity” in this context. Spherical deconvolution models, and the FD / FC / FDC metrics, do not measure or quantify “integrity” in any way. Hopefully re-reading the FBA paper will clarify exactly what these metrics represent.



Dear Rob, thank you very much for the information and your help. I read the FBA paper, however, may I ask you another question: in case of an longitudinal study what would you recommend for a fixel-based analysis, especially considering step 4. Computing group average tissue response functions and step 9. Generate a study-specific unbiased FOD template?

Should I first register each session of each subject internally to an template and then take the averaged as an input for the subject-wise analysis or can I directly take all data including each session of each subject for the fixel-based analysis from the very first beginning?

Thank you for your support!


I believe the generally accepted best approach for longitudinal studies is:

  • Generate an individual template for each subject across sessions.
  • Produce the population template using the set of individual templates as the inputs.
  • For each session:
    • Get a single transformation to population space, based on the composition of (session -> individual template) and (individual template -> population template) transformations
    • Transform and reslice image data in population space using this composed transformation

This is perhaps something that we could provide more details on in the documentation? It seems an approach that a number of people around here are converging on.