Low b-values with increasing and decreasing response function magnitude




I have acquired a diffusion sequence containing a small number of diffusion-weightings at b 500 (n=10) and b 200 (n=6), for free-water estimations outside of MRtrix. The other embedded shells are the now standard b 1000 (n=20) and b 3000 (n=60) weightings.

Upon viewing the response function estimations with the following code:

dwi2response msmt_5tt dwi.mif r5TT.nii wm.txt gm.txt csf.txt -mask brainmask.nii

I noticed that the response function values for the white-matter derived tissue, within each row, are not decreasing in magnitude. The b 1000 shell on later inspection (fourth row), demonstrate values that appear strangely problematic.

To note, the slm-linear option has been passed onto dwipreproc.

However, when I choose only to include the b 1000 and b 3000 shells for analysis, through setting -shells 0,1000,3000, this issue appears to be slightly circumvented:

And when using dhollander:

dhollander, now with only the b 1000 and b 3000 shells, which is more stable:

I presume that the response function estimation is not optimal with a smaller number of vectors? Moreover, if I were to be using this data for MRtrix-alone purposes (i.e. connectome generation and/or fixel-analysis), it’s perhaps best that I remove these sub-b1000 shells?

I did create this diffusion vector set using an older version of gen_scheme, which is illustrated below:


Statistically checking the metrics of the current diffusion vector set does indicate the directions of the sub b 1000 shells are “asymmetric”.

(b=199.99998408333332) [ 6 directions ] 

 Asymmetry of sampling:
    norm of mean direction vector = 0.210423

(b=500.0000467000001) [ 10 directions ]

  Asymmetry of sampling:
    norm of mean direction vector = 0.101242

The full report is attached here

Some basic quality checking of the diffusion signal across the shells:






The first column represents the mean signal in each shell, which is decreasing, the other columns contain the higher order (zonal) harmonic coefficients. You can use shview -response to render your response functions. You should see a sphere that flattens with increasing b-value. See this paper for more details.


Thanks @maxpietsch,

As expected, the harmonic profiles do flatten with increasing b-values:

Although, I believe you slightly misinterpreted my question. My concern is that across the rows within the response function (particularly for the fourth column), the values are not decreasing in magnitude.

To re-demonstrate:

1424.338561611295 0 0 0 0 0
1225.505145912768 -113.8049914665067 5.926890244010404 12.14004544561863 -1.987244138897725 -0.1513613190009629
996.4274607021107 -223.4550228077395 25.92133803121068 -1.706167904058339 2.894367009973647 -2.416315779943908
755.7359056395113 -294.244980880778 60.54861123526116 -9.801876577539458 0.8006653524501117 -3.040617336570455
425.0788579985894 -263.739174182725 128.6349007503337 -45.63362162755784 10.94698263203365 -2.881451573447199

This is unexpected to what I normally obtain with standard one- or- two three shell acquisitions, and oppose what @jdtournier notes within the legacy MRtrix2 documentation.

Perhaps @jdtournier, @ThijsDhollander you could provide some assistance here - albeit I know you guys are very busy with the upcoming workshop!