FBA- Intensities in FOD template and fixel threshold

1

Hi all,

I’m running a fixel-based analysis and have followed the steps in the documentation. The range of intensity values in my FOD template image and template peaks image is larger than expected, so I have to use a higher threshold to obtain the analysis fixel mask. I’m wondering if the range of intensities below is normal, or if it might indicate some other problem?

 mrstats fod_template_image.mif
         volume         mean       median    std. dev.          min          max       count
         [ 0 ]       0.31694            0      0.68983   -0.0527485      3.80367      3538584
         [ 1 ]    0.00366373            0     0.205019     -3.81572      4.23294      3538584
         [ 2 ]    0.00466598            0     0.225624     -4.33142      3.75028      3538584
         [ 3 ]   -0.00482239            0     0.262265     -2.44624      6.00869      3538584
         [ 4 ]  -0.000801694            0     0.230982     -4.61039      4.25817      3538584
         [ 5 ]   -0.00508653            0     0.228359     -3.45561      3.68762      3538584
         [ 6 ]   -0.00131774            0     0.123226     -2.77628      2.80589      3538584
         [ 7 ]  -0.000445164            0     0.123729     -2.38312      2.86549      3538584
         [ 8 ]  -0.000682501            0     0.129962     -2.74054      2.88811      3538584
         [ 9 ]  -0.000126717            0     0.157054     -3.18848       3.7518      3538584
        [ 10 ]   -4.9185e-05            0     0.140962     -1.80616      4.00136      3538584
        [ 11 ]    0.00206355            0     0.152332     -4.02052      3.74121      3538584
        [ 12 ]    0.00161026            0     0.141292     -2.96238      2.84278      3538584
        [ 13 ]  -0.000437462            0     0.128105     -3.03648      3.08545      3538584
        [ 14 ]     0.0104751            0     0.135415     -2.96379      2.68847      3538584
        [ 15 ]  -0.000597185            0    0.0546969     -1.37723      1.36037      3538584
        [ 16 ]   0.000916888            0    0.0573126     -1.13545       1.4431      3538584
        [ 17 ]  -6.38901e-05            0    0.0556259     -1.33657      1.39882      3538584
        [ 18 ]   0.000772601            0    0.0598749     -1.38579      1.43331      3538584
        [ 19 ]   0.000278013            0    0.0654504     -1.71434      1.72772      3538584
        [ 20 ]  -0.000169571            0    0.0667122     -1.76937      2.21278      3538584
        [ 21 ]   -0.00184759            0    0.0619934     -1.11331      1.67619      3538584
        [ 22 ]   0.000203548            0    0.0652077     -2.32206      2.16724      3538584
        [ 23 ]   -0.00010127            0    0.0650521     -1.48514      1.63225      3538584
        [ 24 ]   0.000482334            0    0.0606771     -1.34106       1.3696      3538584
        [ 25 ]    -0.0010348            0    0.0574438     -1.30741      1.34312      3538584
        [ 26 ]   -0.00125646            0    0.0548964     -1.30759       1.3092      3538584
        [ 27 ]   0.000234597            0    0.0588273     -1.42537      1.28587      3538584
        [ 28 ]   0.000242632            0    0.0176822    -0.405565     0.412541      3538584
        [ 29 ]  -0.000114963            0    0.0179523    -0.462116      0.45074      3538584
        [ 30 ]   0.000245818            0    0.0170753    -0.450289     0.466053      3538584
        [ 31 ]  -8.42997e-05            0    0.0179188    -0.411724     0.449423      3538584
        [ 32 ]  -0.000396948            0    0.0175422    -0.427633     0.422398      3538584
        [ 33 ]   0.000244743            0    0.0189737    -0.491464     0.633512      3538584
        [ 34 ]   4.95869e-05            0    0.0188438    -0.597326     0.610163      3538584
        [ 35 ]  -0.000154591            0    0.0189942    -0.627484     0.698862      3538584
        [ 36 ]  -0.000505795            0    0.0187025    -0.397342     0.633377      3538584
        [ 37 ]   0.000313175            0    0.0190023    -0.731432     0.797074      3538584
        [ 38 ]   0.000143332            0    0.0195675    -0.789328     0.500479      3538584
        [ 39 ]    -0.0003236            0    0.0187299     -0.46903     0.499229      3538584
        [ 40 ]  -0.000386468            0    0.0181244    -0.387925      0.41827      3538584
        [ 41 ]   9.72399e-06            0    0.0169208    -0.429927     0.455994      3538584
        [ 42 ]     0.0002302            0    0.0175293    -0.415325     0.428524      3538584
        [ 43 ]   1.46357e-05            0     0.017331    -0.435931      0.41045      3538584
        [ 44 ]   0.000261083            0    0.0184416    -0.365016     0.501498      3538584

fixelstats template_peaks_image.msf
         volume         mean       median    std. dev.          min          max       count
         [ 0 ]       1.22746  0.715628624      1.58389     -1.26308      17.1208      2037132

When I use a threshold of 0.33 as suggested in the documentation, the number of fixels in my analysis mask is very high, compared to when I use a higher threshold of 3.3:

fixelstats analysis_fixel_mask_0.33.msf 
         volume         mean       median    std. dev.          min          max       count
         [ 0 ]        1.3918  0.797295392      1.63646     0.330001      17.1208      1769948

fixelstats analysis_fixel_mask_3.3.msf
         volume         mean       median    std. dev.          min          max       count
         [ 0 ]       5.83222    5.0774188      2.39905          3.3      17.1208       149818

Any help would be appreciated.

Thanks,

Claire

2

Hi Claire,

I checked the stats for my FOD template and they are around 1 order of magnitude lower, the same for the peaks, so has some sense that the threshold that “works” in your case is 3.3, I am not sure right now if that would do some harm to your results.

Have you checked if the values in the individual FODs you use for the template construction are in more or less the same range?

3

Hi @diagiraldo

Yes, the values in the individual FOD images used for the template are similar.

Claire

4

Claire,

The thresholds used for FOD segmentation (which correspondingly determines the number of fixels you get in your template) are based on “typical” FOD sizes. Normally the overall FOD size is sort of “intrinsically” globally normalised based on the overall size of the response function used: If you de-convolve a particular single-fibre voxel signal with a response function defined using that same single-fibre voxel, the FOD should have a total spherical integral of around 1.0 (or an l=0 term of around 1/sqrt(4*pi) ~ 0.28). So it doesn’t matter whether your image intensities are very small or very large: As long as the response function comes from the image data itself, that overall magnitude scaling cancels out during the deconvolution step.

For this reason, my suspicion is that you’ve used the incorrect response function. Say, for instance, you estimate your response function from initial image data, using single-fibre white matter voxels, and it has a spherical integral of e.g. 100; but subsequently to this, you apply dwiintensitynorm to your DWIs, which multiplies all image intensities so that the spherical integral of the diffusion signal in white matter voxels is around 1000. Upon performing deconvolution, your FODs will be 10 times larger than they should be; and as a consequence, spurious peaks and Gibbs ringing normally discarded by the FOD segmentation algorithm will instead be retained.

This I think would explain the effect you’re seeing; so see if you can back-track your steps (and let me know if my hunch was right!)
Rob

5

Hi Rob,

Yes that’s right- I’ve estimated the response function from the original data rather than the intensity normalised data. Thank you for picking up that mistake!

Claire