Are FD measures more affected by susceptibility distortions than FA?

Hello MRtrixers,

I am trying to study a cranial nerve neuropathy using both FD and FA measures. I’m using a multishell protocol (b=1000 and 2500 s/mm²), with AP and PA phase encoding directions. My data includes subjects suffering from the neuropathy and healthy controls for comparison.

First, I denoised my raw data using dwidenoise, and then used FSL topup to correct susceptibility artifacts and FSL eddy for eddy current distortions. I then ran the FBA pipeline to produce FD measures of the cranial nerve (using tract-based fixel masks). For FA, I produced FA maps using the b=1000 s/mm² shell only, and then co-registered the FA maps of the subjects to the population template and used the same tract-based masks to produce FA measures of the cranial nerve.

As expected, I got much better results using FD than FA for the most part. The effect size when comparing healthy cranial nerves to neuropathic ones is much higher using FD compared to FA for most of the length of the cranial nerve. However, due to the close proximity of part of the cranial nerve to a paranasal sinus, FD in that part is much lower than the rest of the nerve, and so is the effect size compared to healthy nerves. In contrast, the part exposed to the susceptibility distortion showed a higher effect size using FA compared to FD.

At first I thought this could due to the fact that I only used the lower b-shell (b=1000) to produce the FA maps, assuming that the higher b-shell would be more affected by susceptibility distortions (please correct me if I’m wrong here). So, I decided to create FA maps from both shells to have a fairer comparison. Again, while the susceptibility-distortion-affected part was noisier using both shells for FA maps, the effect size was still higher using FA compared to FD.

So my question is this: is there something about the way that FD is calculated that would make it inherently more susceptible (for lack of a better word) to suitability distortions compared to measures from the tensor model (such as FA)?? If not, does any one have a possible explanation for my findings?

Any insight would be really appreciated. Thanks!


Hi Joe,

I have explored a lot of patients referred with optic neuropathy (both ischaemic and demyelinating) using MRtrix and single-shell acquisition (b=1000 s/mm2), and tried to do so using multi-shell acquisition more recently.

I can’t really explain differences between FA and FD because I never use FA analysis in this anatomical area (actually in clinical practice, I perform TWI, see this publication.)

However, in my experience it is really hard to set multi-shell DWI parameters that allow to explore the optic nerves canal in which the susceptibility artifacts are very important at high b-values and topup algorithm less efficient.

I have then decided to keep my single shell MR sequence for the optic nerves and to add a second MR multi-shell acquisition for the optic chiasm and optic radiation optimal exploration.

Do you use directional AFD (after tractography) or total AFD coefficient extracted from your optic nerve masks? The former seems more efficient for individual diagnosis yet require to pass through the optic canal :slight_smile: . The latter is not adapted because in a lot of cases, demyelination lesions will produce “crossing fiber” along the optic nerve with a total AFD coefficient close to those of healthy optic nerves.

Don’t hesitate for further information. Best, Arnaud A.