Free water estimation


I would like to extract the free water estimates for several tracts, but I’m not sure if I did it correctly. I have been using the tckmap command with the CSF image generated during the 5TT script as a template and then calculated the stats of the output track-weighted image using mrstats. Is that the correct way of getting the free water estimations?

Cheers, Lena

Hi Lena,

I think there’s confusion surrounding a range of topics here. I’ll try to point you in the right direction, without writing a review article in the process…

“Free water estimate” typically refers to a component of the diffusion model that relates to the amount of free water in the voxel. This is particularly problematic for models that quantify diffusivity, as the presence of any free water at all will significantly bias such estimates. The closest thing we have in MRtrix3 is the “CSF-like” component that can be extracted using multi-tissue CSD: If you have e.g. a white matter voxel that contains a lot of free water, this should be reflected in the estimated signal contribution from the “CSF-like” response function.

This is however a purely voxel-wise estimate, based on the diffusion model. This is not typically a property that is quantified along tracts; this is certainly possible, but it’s necessary to ensure that you have a useful voxel-wise measure before considering tract-specific quantification. Additionally, the 5ttgen script does not use diffusion data in any way; it provides a CSF volume fraction image based on T1 image segmentation, but its estimation is based on vastly different principles to “free water estimation” in the context of diffusion models.

Hope that helps

Hi Rob,

Thanks that is very helpful. I created maps for some tracts with the CSF component I got from the 5TT scripts and then got the stats from that, but from what you’re saying this isn’t quite what I’m after as it isn’t derived from a diffusion image. Do you know if free water estimations will become available in MRtrix at some point?

Cheers, Lena

The multi-shell, multi-tissue (MSMT) variant of constrained spherical deconvolution (CSD) is already available in MRtrix3: the dwi2fod command with specifying msmt_csd as the algorithm. By deriving a response function from CSF voxels and using that in the deconvolution, the resulting density map is the closest analogue we have in MRtrix3 world to what is typically referred to as “free water estimation”. It is however a different model & derivation to the “typical” definition of such, so it’s important to appreciate exactly how these quantities are estimated. I would recommend reading the relevant article and contrasting this technique against other diffusion models that provide “free water estimation”.