Custom loader not used when loading data lazily

[Zhack47]

Is there an existing issue for this?

• I have searched the existing issues

Bug summary

While using a custom loader (in my example .npy loader), if the data is loaded lazily (i.e.: not with subject.load(), but during the application of a transformation for example) Torchio doesnt seem to use the provided thee custom loader, and instead tries to find a fitting SimpleITIKIO reader

This throws an error in this case because no such reader exists for .npy files.

Code for reproduction

import torchio
import numpy as np


def numpy_reader2D_no_affine(path: PosixPath):
    data = np.load(path).astype(np.int32)
    data = data[:, :, 0]
    return data, np.eye(4)

root_dir = "your/npy/data/here" # patient1 ->
                                #            |ct.npy
                                #            |mask.npy
                                # patient2 -> 
                                #            ....

pad_tr = torchio.CropOrPad(patch_size)
for patient in sorted(os.listdir(root_dir)):
    slice_path = os.path.join(root_dir, patient, "ct.npy")
    mask_path = os.path.join(root_dir, patient, "mask.npy")

    l3_slice = torchio.ScalarImage(slice_path, reader=numpy_reader2D_no_affine)
    mask = torchio.ScalarImage(mask_path, reader=numpy_reader2D_no_affine)
    sub = torchio.Subject(ct=l3_slice, gt =mask)
    # sub.load()  # Uncommenting this, the code works properly
    sub = pad_tr(sub)

Actual outcome

Throws error: Unable to determine ImageIO reader for "/mnt/disk_2//Sarco///mask.npy"

Error messages

/home/<user>/.local/lib/python3.11/site-packages/torchio/data/inference/aggregator.py:229: RuntimeWarning: Medical image frameworks such as ITK do not support int64. Casting to int32...
  warnings.warn(message, RuntimeWarning)
Traceback (most recent call last):
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/subject.py", line 309, in check_consistent_space
    self.check_consistent_attribute('spacing')
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/subject.py", line 273, in check_consistent_attribute
    current_attribute = getattr(image, attribute)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/image.py", line 317, in spacing
    _, spacing = get_rotation_and_spacing_from_affine(self.affine)
                                                      ^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/image.py", line 253, in affine
    affine = read_affine(self.path)
             ^^^^^^^^^^^^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/io.py", line 119, in read_affine
    reader = get_reader(path)
             ^^^^^^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/io.py", line 128, in get_reader
    reader.ReadImageInformation()
  File "/home/<user>/.local/lib/python3.11/site-packages/SimpleITK/SimpleITK.py", line 8455, in ReadImageInformation
    return _SimpleITK.ImageFileReader_ReadImageInformation(self)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Exception thrown in SimpleITK ImageFileReader_ReadImageInformation: /tmp/SimpleITK/Code/IO/src/sitkImageReaderBase.cxx:105:
sitk::ERROR: Unable to determine ImageIO reader for "/mnt/disk_2/<user>/<data>/<patient>/mask.npy"

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/<user>/example.py", line 34, in <module>
    sub = pad_tr(sub)
          ^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/transforms/transform.py", line 158, in __call__
    transformed = self.apply_transform(subject)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/transforms/preprocessing/spatial/crop_or_pad.py", line 278, in apply_transform
    subject.check_consistent_space()
  File "/home/<user>/.local/lib/python3.11/site-packages/torchio/data/subject.py", line 320, in check_consistent_space
    raise RuntimeError(message) from e
RuntimeError: As described above, some images in the subject are not in the same space. You probably can use the transforms ToCanonical and Resample to fix this, as explained at https://github.com/fepegar/torchio/issues/647#issuecomment-913025695

Expected outcome

No error

System info

Platform:   Linux-5.15.0-74-generic-x86_64-with-glibc2.35
TorchIO:    0.18.78
PyTorch:    1.13.1+cu117
SimpleITK:  2.1.1.2 (ITK 5.2)
NumPy:      1.22.0
Python:     3.10.6 (main, Mar 10 2023, 10:55:28) [GCC 11.3.0]

[fepegar]

Thanks for reporting, @Zhack47. This should be fixed in v0.18.92.

@all-contributors please add @Zhack47 for bug

[allcontributors]

@fepegar

I couldn't determine any contributions to add, did you specify any contributions?
Please make sure to use valid contribution names.

I've put up a pull request to add @Zhack47! 🎉