Apparent fiber density vs Neurite density index (NODDI)

Hello MRtrixers,

I’m using both Apparent fiber density (AFD) and NODDI to study WM degeneration in a group of patients compared to healthy controls. I found a significant loss of AFD in the patient group, while the Neurite density index (NDI) produced by NODDI showed no difference (not even close, so it’s not a matter of statistical power). More interestingly, the orientation dispersion index (ODI) showed a significant increase in the patient group, which was quite surprising,

I know AFD and NDI rely on different models, but it’s my understanding is that they both represent the intra-axonal compartment, correct? If so, does anyone have any idea what could be the cause of the lack of NDI loss that I’m finding? And is it perhaps related somehow to the increase of ODI?

Any insight/opinions would be appreciated. Thanks!


Hello again,

I have yet to find an explanation to my findings. I went through as much literature as I could get my hands on, but didn’t find anything useful. Any suggestion would be appreciated. Thanks!


it’s my understanding is that they both represent the intra-axonal compartment, correct?

They both ideally represent the intra-axonal compartment. But the two derivations have different assumptions, and will respond differently to different breaches of those assumptions. For instance the proportionality of AFD to the intra-cellular compartment is strongly dependent on a sufficiently high b-value, such that the extra-cellular component doesn’t contribute to the DWI signal, let alone any modeling.

One detail absent from your reporting is the domain of quantification. NDI is a voxel-wise measure, whereas AFD is (typically) a fixel-wise measure. If you were to be quantifying the voxel-wise total AFD, and subjecting such to precisely the same statistical analysis as NDI, then I would focus attention on the differences between the two models; if however you are comparing voxel-wise NDI to fixel-wise AFD, it may actually be the domain of quantification that is driving the difference.

For instance, imagine two voxel, both of which contain two fibre populations at 90 degrees to one another, but the volume fractions differ between the two voxels:

  1. Fibre 1 = 40%, Fibre 2 = 40%, extra-cellular = 20%

  2. Fibre 1 = 10%, Fibre 2 = 70%, extra-cellular = 20%

With AFD / FBA, you would ideally report differences in both fibre 1 and fibre 2.

With NODDI, you would report no change in NDI. There would however be a difference in ODI, since in voxel 2 the fibre density is pretty coherently aligned whereas in voxel 1 it is not.

Even if not the answer, maybe it at least provides some ideas for how to interrogate the data further.



This is a bit surprising for me, as a few previous studies have shown that NDI is quite sensitive to neurodegenerative processes including some of my own work. Could you please give a more detailed description of your data acquisition and your sample description? Am happy to help look further with this issue.