FBA - problems with masks

[katoherregods]

Hi Mrtrixers!

For my master thesis, I am performing the MSMT-FBA analysis exactly as described in Mrtrix-readthedocs, but

I noticed that my masks include a LOT of non-brain regions.

This is the case for both my binary, upsampled brain masks created for each subject (Step 6, dwi2mask), as well as the study-specific FOD template. (Step 9, population_template), the mask warped into template space (Step 11.1, mrtransform) and template mask (Step 11.2, mrmath). In addition, the binary template mask also contains some holes.

For my template analysis fixel mask (Step 12, fod2fixel), the problem is two fold; genuine whiter matter voxels seems to be excluded in the anterior and posterior part of the brain, but a large redundant part is also found at the bottom.

I think the problem starts in the first binary masks, and gets transmuted from their. How do I make the first upsampled brain masks fit better? How do I make the fixel mask fit better? Are additional steps needed to improve the other masks?

As for the upsampled brain mask images(step 6), it was mentioned that a generous mask was best, as regions that did not get included at this point, would not be used for FOD generation. However, for bias field correction and intensity normalisation, this same mask is used, but this time I am adviced to use a conservative mask (Step 9). How can these two pieces of advice be combined? Can I still use the same upsampled brain masks generated in step 6 for performing step 9?

Thanks a lot in advance!

Kato

Some additional information:

I preprocessed my data using a QSIprep container.

bval’s:

0 0 2500 1650 2500 850 1650 2500 1800 1650 850 1300 1650 2500 2500 1650 2500 850 0 1650 1850 2500 1700 850 850 1150 2500 2500 1650 2500 850 1650 1600 1600 1650 0 2500 850 1650 2500 2500 1650 2500 850 1650 2500 2500 1650 1600 850 1650 2500 0 2500 1650 2500 850 1650 1600 2500 1650 850 2500 1650 1600 2500 1650 1600 850 0 1650 2500 2500 1650 850 2500 1650 2500 2500 1650 850 2500 1650 1450 2500 1650 0 850 2500 1650 2500 2500 1650 850 2500 1650 2500 2500 1650 2500 850 1650 2500 0

mrinfo -size directions.mif:

864 964

[aknodt]

i’ve been dealing with a similar issue that i’ve posted about here:

Question about FBA sensitivity to mask quality

I am new to FBA but seem to have successfully executed the analysis for most of the subjects in my dataset. One of the things I am QC’ing on everyone is the quality of the mask, and I’ve found that quite a few seem to be problematic esp according to the note in Section 8 of the tutorial (”Joint bias field correction and intensity normalisation”) stating that mtnormalise results can be sensitive to masks that contain non-brain voxels. Probably > 10% of my subjects have masks that substantially …

i unfortunately don’t have a solution for you, but maybe you could also find some improvement using synthstrip as an alternative masking strategy - it helped for some but not all of my scans. it seems like improving the masking function in mrtrix might be an active area of development. but like you i’m also still wondering what the implications of the poor masks are, and for now i am just excluding those scans where the masks aren’t very good.

[WilliamFCB]

Hi, it is important that the mask fits what you want to do (more restrictive/more inclusive). Personally, I used SynthStrip to create masks from MeanBO images that qsiprep generates. If you used T1 data in your QSIprep preprocessing, the mask is likely derived from the T1 and might include cerebellum tissue not necessarily covered by your DWI. You can also generate FA images and derive brain masks from those. Depending on your use case, you can use MRtrix3 “maskfilter” ( maskfilter — MRtrix3 3.0 documentation ) to erode or dilate masks, if necessary. It is important to check the quality of the mask (and your data in general). I use FSL, slicesdir, to generate some htmls for a quick evaluation.