Dwi2fod with FA as mask

Hello all,

Could someone please tell me if it would be beneficial to provide a thresholded binarized mask built from FA map to dwi2fod?

Thank you,

Any idea about my question above?

Nope, no benefit that I can think of. The only reason to use a mask for dwi2fod is to reduce processing time by avoiding needless (but otherwise harmless) computation in non-brain regions. However, using a thresholded FA mask will likely mean genuine brain, and worse, white matter voxels will be excluded – it’ll leave holes in your output FOD images.

Thanks, Donald!

I’m using single shell data (+ b0) in the msmt_csd scheme so I’m only getting the wmfod, csf, and GM and as a consequence the wmfod is overestimated (includes the GM tissue?). So, I was hoping by providing a FA map I could somehow help the algorithm to not include some of the GM.
BTW, how can I create a binary mask of my FA map in way that FA values between 0.1 and 0.9 maps to 1 and the rest to 0?


Not sure exactly what you mean here, but what we’d recommend with single shell data is to use dwi2fod msmt_csd to produce a 2 tissue decomposition using the WM and CSF responses only, ignoring the GM (see e.g. this post). There’s only 2 b-values, you can’t really estimate more than 2 components with multi-tissue CSD. It does mean that the GM component will end up contaminating both the WM and CSF components to some extent, but it’s still cleaner than single-shell CSD.

The processing is performed independently per voxel. All the mask does is specify which voxels get processed. So you’ll get the same results as just setting those voxels to zero after the dwi2fod call – it’s certainly not going to affect the results in those voxels that are being processed.

Lots of ways:

mrthreshold fa.mif -abs 0.1 mask1.mif
mrthreshold fa.mif -abs 0.9 -invert mask2.mif
mrcalc mask1.mif mask2.mif -mult mask.mif

Or using mrcalc:

mrcalc fa.mif 0.1 -gt mask1.mif
mrcalc fa.mif 0.9 -lt mask2.mif
mrcalc mask1.mif mask2.mif -mult mask.mif

Or better, a one-liner using mrcalc:

mrcalc fa.mif 0.1 -gt fa.mif 0.9 -lt -mult mask.mif

Thanks for your clarification, Donald!

For reference, other considerations to be careful about (e.g. with mtnormalise and other steps in mind): FBA Single Shell - High GM FOD values

It might not really have been mentioned before as well, but also be careful what masks you give to mtnormalise if you’d run 2- or even 3-tissue CSD only in an FA>… mask. You may cut out most if not all of the areas that have contributions of the other tissue types (and more, e.g. introduce holes). That’s going to be a problem that’ll undermine mtnormalise's accuracy, potentially a lot. mtnormalise thrives upon the info of all tissue types to do its job well (it’s more complicated than that; but it’s the simplest way I can put it).

Thanks, Thijs.

1 Like