Affine transform between torchio.ScalarImage and nibabel.load do not match

🐛Bug I am using the following NIfTI volume that is part of a dataset that I am using (don’t worry, this is public data and not confidential). When you use torchio.ScalarImage and examine the affine transform after it has been loaded, it is not the same as if you were to use nibabel.load. This causes havoc when I use the ToCanonicaltransform because the volume is not oriented correctly.

To reproduce

Please download the file from the link above. Once you do, do the following:

import torchio as tio
import nibabel as nib

vol_path = 'img0009.nii.gz'
img_s = tio.ScalarImage(vol_path)
print(img_s.affine)
# array([[-0.97656202,  0.        ,  0.        ,  0.        ],
#       [ 0.        , -0.97656202,  0.        ,  0.        ],
#       [ 0.        ,  0.        ,  2.5       ,  0.        ],
#       [ 0.        ,  0.        ,  0.        ,  1.        ]])
img_n = nib.load(vol_path)
print(img_n.affine)
# array([[  -0.97656202,    0.        ,    0.        ,  249.51159692],
#       [   0.        ,    0.97656202,    0.        , -249.51159692],
#       [   0.        ,    0.        ,    2.5       , -185.        ],
#       [   0.        ,    0.        ,    0.        ,    1.        ]])

I am using Nibabel version 3.2.1 and torchio version 0.18.29.

Expected behavior

I would expect the affine transforms between the calls to torchio.ScalarImage and nib.load to be exactly the same. Why is the affine transform modified?

Actual behavior

The affine transforms are different.

System info

Platform: Linux-4.15.0-91-generic-x86_64-with-debian-buster-sid
TorchIO:  0.18.29
PyTorch:  1.7.1
NumPy:    1.19.2
Python:   3.7.3 (default, Mar 27 2019, 22:11:17)
[GCC 7.3.0]

[rayryeng]

I got it! Thank you so much! I shall close this.

[fepegar]

No worries! Actually, TorchIO uses NiBabel only if SimpleITK fails to read the file.