LoadImage can not read .mgz files

[Spenhouet]

Is your feature request related to a problem? Please describe.
If one wants to work with Freesurfer data it is necessary to read .mgz files.
Nibabel can do so.

Describe the solution you'd like
One step is to add the .mgz extension to the suffixes defined for the NibabelReader.

suffixes: Sequence[str] = ["nii", "nii.gz"] # <- add mgz here

I tested this by monkey patching NibabelReader.verify_suffix:

from monai.data.utils import is_supported_format
from monai.data.image_reader import NibabelReader


def verify_suffix(self, filename: Union[Sequence[str], str]) -> bool:
    """
    Verify whether the specified file or files format is supported by Nibabel reader.

    Args:
        filename: file name or a list of file names to read.
            if a list of files, verify all the suffixes.

    """
    suffixes: Sequence[str] = ["nii", "nii.gz", "mgz"]
    return is_supported_format(filename, suffixes)


NibabelReader.verify_suffix = verify_suffix

But there seem to be more adjustments needed. With this change I get the following exception:

~\AppData\Local\Programs\Python\Python37\lib\site-packages\monai\data\image_reader.py in read(self, data, **kwargs)
    363             img = nib.load(name, **kwargs_)
--> 364             img = correct_nifti_header_if_necessary(img)
    365             img_.append(img)

~\AppData\Local\Programs\Python\Python37\lib\site-packages\monai\data\utils.py in correct_nifti_header_if_necessary(img_nii)
    489     """
--> 490     dim = img_nii.header["dim"][0]
    491     if dim >= 5:

~\AppData\Local\Programs\Python\Python37\lib\site-packages\nibabel\freesurfer\mghformat.py in __getitem__(self, item)
    495             return self._header_data[item]
--> 496         return super(MGHHeader, self).__getitem__(item)
    497 

~\AppData\Local\Programs\Python\Python37\lib\site-packages\nibabel\wrapstruct.py in __getitem__(self, item)
    313         """
--> 314         return self._structarr[item]
    315 

ValueError: no field of name dim

Describe alternatives you've considered
One can convert the .mgz files to .nii.gz beforehand like this:

img = nib.load(f"{filename}.mgz")
nib.save(img , f"{filename}.nii.gz")

[Nic-Ma]

Hi @Spenhouet ,

May I know the error if reading the .mgz directly with LoadImage? I think ITK should support it?

Thanks.

[Spenhouet]

Hi @Nic-Ma,

thanks for the hint. I did not know that ITK is the catch all for loading image files.
I did not have ITK installed and therefore got:

RuntimeError: can not find suitable reader for this file: path\to\my\file\orig.mgz.                 Please install dependency libraries: (nii, nii.gz) -> Nibabel, (png, jpg, bmp) -> PIL,                 (npz, npy) -> Numpy, others -> ITK. Refer to the installation instruction:                 https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies.

Nonetheless, I installed ITK now and still can not load the image.

~\AppData\Local\Programs\Python\Python37\lib\site-packages\monai\data\image_reader.py in read(self, data, **kwargs)
    195 
--> 196             img_.append(itk.imread(name, **kwargs_))
    197         return img_ if len(filenames) > 1 else img_[0]

~\AppData\Local\Programs\Python\Python37\lib\site-packages\itk\support\extras.py in imread(filename, pixel_type, fallback_only, imageio)
    956     else:
--> 957         reader = template_reader_type.New(**kwargs)
    958     reader.Update()

~\AppData\Local\Programs\Python\Python37\lib\site-packages\itk\support\template_class.py in New(self, *args, **kwargs)
    660             return self._NewImageReader(
--> 661                 itk.ImageFileReader, False, "FileName", *args, **kwargs
    662             )

~\AppData\Local\Programs\Python\Python37\lib\site-packages\itk\support\template_class.py in _NewImageReader(TemplateReaderType, increase_dimension, primaryInputMethod, *args, **kwargs)
    160             raise RuntimeError(
--> 161                 f"Could not create IO object for reading file {inputFileName}" + msg
    162             )

RuntimeError: Could not create IO object for reading file path\to\my\file\orig.mgz

[Nic-Ma]

Hi @Spenhouet ,

Could you please help share your test program for further analysis?
@thewtex Do you have any idea about this ITK error?

Thanks in advance.

[Spenhouet]

Sure, this is the minimal reproduction:

from monai.transforms import LoadImage

LoadImage()(r'C:\Users\path\to\my\file\orig.mgz')

I can't share my input file.

But I would propose to simply enable the NibabelReader to handle .mgz files.

The following patch works:

from typing import List, Sequence, Union

import nibabel as nib
from monai.data.image_reader import NibabelReader, has_nib
from monai.data.utils import (correct_nifti_header_if_necessary, is_supported_format)
from monai.transforms import LoadImage
from monai.utils import ensure_tuple
from nibabel.nifti1 import Nifti1Image


def verify_suffix(self, filename: Union[Sequence[str], str]) -> bool:
    """
    Verify whether the specified file or files format is supported by Nibabel reader.

    Args:
        filename: file name or a list of file names to read.
            if a list of files, verify all the suffixes.

    """
    suffixes: Sequence[str] = ["nii", "nii.gz", "mgz"]  # <- Added 'mgz' here
    return has_nib and is_supported_format(filename, suffixes)


def read(self, data: Union[Sequence[str], str], **kwargs):
    """
        Read image data from specified file or files.
        Note that the returned object is Nibabel image object or list of Nibabel image objects.

        Args:
            data: file name or a list of file names to read.
            kwargs: additional args for `nibabel.load` API, will override `self.kwargs` for existing keys.
                More details about available args:
                https://github.com/nipy/nibabel/blob/master/nibabel/loadsave.py

        """
    img_: List[Nifti1Image] = []

    filenames: Sequence[str] = ensure_tuple(data)
    kwargs_ = self.kwargs.copy()
    kwargs_.update(kwargs)
    for name in filenames:
        img = nib.load(name, **kwargs_)
        if not isinstance(img, Nifti1Image):   # <- added this line
            img = Nifti1Image.from_image(img)  # <- added this line
        img = correct_nifti_header_if_necessary(img)
        img_.append(img)
    return img_ if len(filenames) > 1 else img_[0]


NibabelReader.verify_suffix = verify_suffix
NibabelReader.read = read

LoadImage()(r'C:\Users\path\to\my\file\orig.mgz')

It is only a small change which allows Nibabel to convert the MGHImage object to a Nifti1Image object.

This is basically the fix:

if not isinstance(img, Nifti1Image):
    img = Nifti1Image.from_image(img)

Why is Nifti1Image over Nifti2Image used?

[Nic-Ma]

Hi @Spenhouet ,

I tried to run the ITK reader, maybe it because the image file is unavailable?
Could you please help confirm with below command:

>>> import itk
>>> 
>>> itk.imread(r'path/to/orig.mgz')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/conda/lib/python3.8/site-packages/itk/support/extras.py", line 957, in imread
    reader = template_reader_type.New(**kwargs)
  File "/opt/conda/lib/python3.8/site-packages/itk/support/template_class.py", line 660, in New
    return self._NewImageReader(
  File "/opt/conda/lib/python3.8/site-packages/itk/support/template_class.py", line 160, in _NewImageReader
    raise RuntimeError(
RuntimeError: Could not create IO object for reading file path/to/orig.mgz
The file doesn't exist. 
Filename = path/to/orig.mgz

[wyli]

I feel if the user specifies a reader in LoadImage, there's no need to do the filename extension checks. I'll submit a PR to show the idea

[Nic-Ma]

Hi @wyli ,

I see your idea, that makes sense, we should only check extension in "auto" mode.

Thanks.

[Spenhouet]

Hi @Nic-Ma,

import itk

itk.imread(r'C:\Users\path\to\my\file\orig.mgz')

This gives me the same error as with LoadImage:

RuntimeError: Could not create IO object for reading file C:\Users\path\to\my\file\orig.mgz

With the fix for the NibabelReader above I can load the same file so it is definitely not that.

@wyli @Nic-Ma What you do think about the proposed adjustment to the NibabelReader?

[Nic-Ma]

I think @wyli wanted to submit a PR to unlock the extension check if explicitly providing reader, then you can just set NibabelReader for your case.
But I am still don't understand why ITK can't read it, CC @thewtex .

Thanks.

[Spenhouet]

I tested the fix with LoadImaged(FILE_KEYS, "NibabelReader") and can confirm that it solved my issue.

[Spenhouet]

Thanks @wyli for the implementation. :)