search

MRtrix3 / mrtrix3

MRtrix3 provides a set of tools to perform various advanced diffusion MRI analyses, including constrained spherical deconvolution (CSD), probabilistic tractography, track-density imaging, and apparent fibre density
http://www.mrtrix.org
Mozilla Public License 2.0
292 stars 179 forks source link

hcpmmp1_ordered.txt update #2849

Open arkiev opened 7 months ago

arkiev commented 7 months ago

The current hcpmmp1_ordered.txt LUT lists abbreviated structures that don't match with the contents of FreeSurferSGM.txt, and as such, labelsgmfix returns an error. This relates to the file located at /opt/mrtrix3/latest/share/mrtrix3/labelconvert/hcpmmp1_ordered.txt

As a workaround, I locally update the labels of the following structures to: 

361  Left-Cerebellum  230 148  34 255
362  Left-Thalamus-Proper      0 118  14 255
363  Left-Caudate     122 186 220 255
364  Left-Putamen     236  13 176 255
365  Left-Pallidum     12  48 255 255
366  Left-Hippocampus 220 216  20 255
367  Left-Amygdala    103 255 255 255
368  Left-Accumbens-area   255 165   0 255
369  Left-VentralDC   165  42  42 255
370  Right-Cerebellum  230 148  34 255
371  Right-Thalamus-Proper      0 118  14 255
372  Right-Caudate     122 186 220 255
373  Right-Putamen     236  13 176 255
374  Right-Pallidum     13  48 255 255
375  Right-Hippocampus 220 216  20 255
376  Right-Amygdala    103 255 255 255
377  Right-Accumbens-area   255 165   0 255
378  Right-VentralDC   165  42  42 255

These changes need to be made after every build...Any chance of updating this at the source?

Lestropie commented 6 months ago

Before making any change to such raw data, I would need to be confident that doing so isn't going to break some other workflow.

Current contents of share/mrtrix3/FreeSurferSGM.txt:

1  Left-Thalamus-Proper
2  Left-Caudate
3  Left-Putamen
4  Left-Pallidum
5  Left-Accumbens-area

6  Right-Thalamus-Proper
7  Right-Caudate
8  Right-Putamen
9  Right-Pallidum
10 Right-Accumbens-area

11 Left-Hippocampus
12 Left-Amygdala
13 Right-Hippocampus
14 Right-Amygdala

Current contents of share/mrtrix3/labelconvert/fs_default.txt:

36    L.TH    Left-Thalamus                   0   118 14  255
36    L.TH    Left-Thalamus-Proper            0   118 14  255
37    L.CA    Left-Caudate                    122 186 220 255
38    L.PU    Left-Putamen                    236 13  176 255
39    L.PA    Left-Pallidum                   12  48  255 255
40    L.HI    Left-Hippocampus                220 216 20  255
41    L.AM    Left-Amygdala                   103 255 255 255
42    L.AC    Left-Accumbens-area             255 165 0   255

43    R.TH    Right-Thalamus                  0   118 14  255
43    R.TH    Right-Thalamus-Proper           0   118 14  255
44    R.CA    Right-Caudate                   122 186 220 255
45    R.PU    Right-Putamen                   236 13  176 255
46    R.PA    Right-Pallidum                  13  48  255 255
47    R.HI    Right-Hippocampus               220 216 20  255
48    R.AM    Right-Amygdala                  103 255 255 255
49    R.AC    Right-Accumbens-area            255 165 0   255

Current contents of share/mrtrix3/labelconvert/hcpmmp1_ordered.txt (added in #1289):

361  L_Cerebellum  230 148  34 255
362  L_Thalamus      0 118  14 255
363  L_Caudate     122 186 220 255
364  L_Putamen     236  13 176 255
365  L_Pallidum     12  48 255 255
366  L_Hippocampus 220 216  20 255
367  L_Amygdala    103 255 255 255
368  L_Accumbens   255 165   0 255
369  L_VentralDC   165  42  42 255
370  R_Cerebellum  230 148  34 255
371  R_Thalamus      0 118  14 255
372  R_Caudate     122 186 220 255
373  R_Putamen     236  13 176 255
374  R_Pallidum     13  48 255 255
375  R_Hippocampus 220 216  20 255
376  R_Amygdala    103 255 255 255
377  R_Accumbens   255 165   0 255
378  R_VentralDC   165  42  42 255

The header of this file also states:

This tableprovides parcel names and colours as they are provded from e.g. the following command: mri_aparc2aseg --old-ribbon --annotation glasser --s hmsc001-01 --o glasser.mgz

Also these strings are identical to those in share/mrtrix3/labelconvert/hcpmmp1_original.txt.

If my recollection is right, there is no suitable lookup table provided to accompany the output of the command used for generating the parcellation image. As such, MRtrix3 embeds both a table suitable for interpreting its output, and a target reordering, to be used as input and output LUTs for labelconvert respectively.

So I see two potential solutions:

  1. Those parcels are renamed in both share/mrtrix3/labelconvert/hcpmmp1_original.txt and share/mrtrix3/labelconvert/hcpmmp1_ordered.txt (renaming in only the former would break the labelconvert step). Potential disadvantages of this approach:

    1. Any other software that is predicated on the current naming of these nodes within these particular files would fail.
    2. The names of these parcels would no longer follow the abbreviated convention used throughout the rest of the file. This might be detrimental for any kind of visualisation or data interrogation.

  2. Modify the labelsgmfix code, and/or the contents of share/, to make that script applicable to data where the HCPMMP1 parcellation has been applied. What might actually work straight out of the box here based on a cursory inspection would be to instead modify share/mrtrix3/labelsgmfix/FreeSurferSGM.txt to contain both conventions; ie.:

    1  Left-Thalamus-Proper
1  L_Thalamus
2  Left-Caudate
2  R_Thalamus
3  Left-Putamen
3  L_Putamen
4  Left-Pallidum
4  L_Pallidum
5  Left-Accumbens-area
5  L_Accumbens

6  Right-Thalamus-Proper
6  R_Thalamus
7  Right-Caudate
7  R_Caudate
8  Right-Putamen
8  R_Putamen
9  Right-Pallidum
9  R_Pallidum
10 Right-Accumbens-area
10 R_Accumbens

11 Left-Hippocampus
11 L_Hippocampus
12 Left-Amygdala
12 L_Amygdala
13 Right-Hippocampus
13 R_Hippocampus
14 Right-Amygdala
14 R_Amygdala

    MRtrix3 permits LUTs to have multiple labels for one index in certain contexts; this is one such way in which this can be exploited.

My preference is for the second approach, provided it works for both HCPMMP1 and other data.

Given you've already got one posthoc hack in place, can you see if this alterative posthoc hack works?

arkiev commented 6 months ago

Thanks for looking into this, and for the suggestions.

Here's what I have done: return /opt/mrtrix3/latest/share/mrtrix3/labelconvert/hcpmmp1_ordered.txt to the default file, updated /opt/mrtrix3/latest/share/mrtrix3/labelsgmfix/FreeSurferSGM.txt to contain both conventions, re-ran my code...

This change fixed the problem with hcpmmp1 parcellation, and did not cause errors with destrieux, desikan, Yeo7 or Yeo17...so perhaps let's continue using FreeSurferSGM.txt that contains both conventions going forward?

Lestropie commented 6 months ago

I'm happy with that. Even better would be to add some test data to prove that the fix yields a sensible result and doesn't break any existing test... but I won't put you through that. I'm just neck deep in CI of late.