TST+FIX: LTA conversions

[mgxd]

No description provided.

[oesteban]

I think it won't change your tests, but I think this line has source and destination swapped:
https://github.com/poldracklab/nitransforms/blob/45a0a2fe4268714b0f7237bab65cb084d806fa68/nitransforms/io/lta.py#L243

That line would work okay if xform['m_L'] were the ras2ras matrix and you wanted to calculate the vox2vox matrix from it. However, we want to do the opposite:

• Let R be a matrix representing a ras2ras transform and V the matrix representing the equivalent vox2vox transform.
• Let Asrc and Adst be the affines of source and destination, respectively.

Then, R = inv(Adst) . V . Asrc.

So, if we want to extract V from R, we need to multiply the inverses of each side:

            M = dst.as_affine().dot(xform['m_L'].dot(np.linalg.inv(src.as_affine())))

[mgxd]

yeah, i was just looking at how LTA convert implements and you got it
https://github.com/freesurfer/freesurfer/blob/933dc7794c15243e7f0370e912c6e877b87eda66/utils/transform.cpp#L3707

Even changing the algorithm, I'm still not sure why it is not working in my example...

[oesteban]

Can you update the line with my suggestion?

[mgxd]

Okay - I decided to use the affine-RAS.fs.lta as the baseline (RAS2RAS), and convert it to VOX2VOX using lta_convert. There is still a slight discrepancy, albeit much smaller than before.

In [3]: v2v_m                                                                                                                                                                                                                               
Out[3]: 
array([[ 9.9999899e-01, -9.9999935e-04,  9.9999981e-04,  3.9999959e+00],
       [ 1.4049364e-03,  6.2160885e-01, -7.8332651e-01,  1.9999948e+00],
       [ 1.6171722e-04,  7.8332722e-01,  6.2160969e-01, -9.9998468e-01],
       [ 0.0000000e+00,  0.0000000e+00,  0.0000000e+00,  1.0000000e+00]],
      dtype=float32)

In [4]: r2r_m                                                                                                                                                                                                                               
Out[4]: 
array([[ 9.9999899e-01, -9.9999935e-04,  9.9999981e-04,  4.0000000e+00],
       [ 1.4049363e-03,  6.2160885e-01, -7.8332651e-01,  2.0000000e+00],
       [ 1.6171722e-04,  7.8332716e-01,  6.2160963e-01, -1.0000000e+00],
       [ 0.0000000e+00,  0.0000000e+00,  0.0000000e+00,  1.0000000e+00]],
      dtype=float32)

[codecov-io]

Codecov Report

Merging #36 into master will increase coverage by 2.15%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #36      +/-   ##
==========================================
+ Coverage   91.92%   94.08%   +2.15%     
==========================================
  Files           8        8              
  Lines         570      575       +5     
  Branches       75       77       +2     
==========================================
+ Hits          524      541      +17     
+ Misses         30       18      -12     
  Partials       16       16

Flag	Coverage Δ
#unittests	94.08% <100%> (+2.15%)	⬆️

Impacted Files	Coverage Δ
nitransforms/linear.py	78.33% <ø> (ø)	⬆️
nitransforms/io/lta.py	98.57% <100%> (+8.94%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 455f337...6733cba. Read the comment docs.

[oesteban]

Left a few comments, but this is looking great. Thanks a lot!

[oesteban]

Only one minor nitpick