Controlling for FWF in white matter analyses

[ranesh-neuro01]

Hi all,

We are looking at white matter tracts using Human Connectome Project 7T data. We wanted to look at NODDI metrics along each tract, for which we are taking an AFQ-style approach where we split the tract into 100 nodes and extract NODDI metrics at each node.

Given that this is a sample of healthy young adults, and we don't really expect edema or neuroinflammation, we are considering restricting our analyses to just NDI and ODI. Furthermore, given the curvature of our tracts as well as it passing through regions of crossing fibers, and proximity to ventricles, FWF values may be inflated along these tracts, thus indirectly biasing NDI and ODI estimations. I wonder if it could be worth controlling for FWF along the tract to obtain fraction specific indices of tract microstructure we are interested in (ODI and NDI), however I worry this may be statistically redundant since FWF also shares mathematical components as the other NODDI metrics.

Alternatively would you recommend we drop subjects who have a certain number of nodes that exceed a cutoff value for FWF? Ideally though I would like to retain as many subjects as possible, so I am not the most keen on this idea.

I am curious to see what your thoughts on this matter are. Do you have any other suggestions we could implement to obtain optimal estimates of microstructure along tracts?

[daducci]

Indeed, NODDI models isotropic contributions (i.e. free water) as an independent compartment, but it’s also true that the more you get closer to the ventricles, the model degenerates in particular for NDI values (which you are actually interested in). For instance, if you fit NODDI in a pure CSF voxel, then the NDI value can be any value, right?

I am not an expert in statistical analyses, but what you suggest seems very fair to me. I would, however, not discard subjects based on FWF values, as this should, in principle, be controlled for by using FWF values as covariates, no?