dicom nrrd nifti spacing, origin mismatch.

banikr · August 4, 2021, 3:06pm

Hello I am working with dicom and nrrd image file formats. Usually nrrd files are segmentations and in different spacing and origin compared to the dicoms. So my workflow is:

resample the nrrd file on dicom to match dicom’s spacing and origin and other image properties.
Then perform some morphological operation on the resampled nrrd file.
convert both dicom and resampled nrrd to nifti format maintaining the properties.

For these 3 steps I am using the following:

reader = sitk.ImageSeriesReader()
dicom_names = reader.GetGDCMSeriesFileNames(dcmCasePath)
reader.SetFileNames(dicom_names)
dicom = reader.Execute() 

nrrd = sitk.ReadImage(nrrdFilePath)

use the following function to resample nrrd on dicom:

def Realign(self, image, reference_image, def_value=0.0):
        """
        :param image: image to resample or align(.nrrd files)
        :param reference_image: .dcm files
        :param def_value: values to fill the resampled image
        :return: aligned image that matches reference image origin, spacing etc.
        """
        return sitk.Resample(image,
                             reference_image,
                             sitk.Transform(),
                             sitk.sitkLinear,
                             def_value,
                             reference_image.GetPixelID())
data = [dicom, nrrd]
alignedNrd= Realign(data[1], data[0])

Then I perform some morphological operations on NumPy-based aligned image.

SegBN = np.zeros_like(sitk.GetArrayFromImage(alignedNrd))
SegBN[np.where(sitk.GetArrayFromImage(alignedNrd) > 0)] = 1

Finally saving the image in nifti:

nrd2niiPath = os.path.splitext(nrdpath)[0]+'_{}'.format(ig)+'.nii.gz'
sitk.WriteImage(sitk.GetImageFromArray(SegBN), nrd2niiPath)
dcm2niiPath = os.path.join(dcmPath, os.path.basename(dcmPath)) +'.nii.gz'
sitk.WriteImage(dicom, dcm2niiPath)

But the spacing and origin in the nifti images are totally different. Despite the nrrd image was resampled in reference to the dicom.

To check:

dcmimage.GetSpacing(), alignedNrrd.GetSpacing()
((0.3613280058, 0.3613280058, 0.625), (0.3613280058, 0.3613280058, 0.625))

dcmimage.GetOrigin(), alignedNrrd.GetOrigin()
((-69.19999695, -109.8000031, -240.375),
 (-69.19999695, -109.8000031, -240.375))

dcmimage.GetSize(), alignedNrrd.GetSize()
((512, 512, 224), (512, 512, 224))

imgnb.get_affine(), segnb.get_affine()
(array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]),
 array([[-1.,  0.,  0., -0.],
        [ 0., -1.,  0., -0.],
        [ 0.,  0.,  1.,  0.],
        [ 0.,  0.,  0.,  1.]]))

As it can be seen the nifti formats are in different spacings.
To visualize the nifti image and segmentation separately in ITKsnap:

For some reason, the segmentation is isotropic and completely in different spacing and origin, despite the image size are the same.

What is happening after the morphological operation or saving niftis that changes the image origin, spacing and all?

dchen · August 4, 2021, 3:14pm

Converting your image to NumPy arrays loses all the metadata information.

dzenanz · August 4, 2021, 3:16pm

When you create SegBN, you lose image metadata. Try this:

SegBN = sitk.GetArrayFromImage(alignedNrd)
SegBN[np.where(sitk.GetArrayFromImage(alignedNrd) > 0)] = 1

banikr · August 4, 2021, 4:34pm

It didn’t solve the spacing problem.

>> imgnb.affine, segnb.affine
(array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]),
 array([[-1.,  0.,  0., -0.],
        [ 0., -1.,  0., -0.],
        [ 0.,  0.,  1.,  0.],
        [ 0.,  0.,  0.,  1.]]))

Additionally, I can’t change the morphological operations. I have to create a binary mask, and the one you suggested creates a map with 3 values(-1, 0, 1).

banikr · August 4, 2021, 4:35pm

So are there ways to perform morphological operations without converting resampled nrrd images to NumPy arrays?

dzenanz · August 4, 2021, 4:35pm

Perhaps than copy metadata after you create the segmentation:

SegITK = sitk.GetImageFromArray(SegBN)
SegITK.CopyInformation(alignedNrd)
...
sitk.WriteImage(SegITK, nrd2niiPath)

dchen · August 4, 2021, 4:46pm

It looks like you’re doing a binary threshold on alignedNrd, right?

If so, you should be able to do something like this:

SegBN = (alignedNrd > 0)

This does the same thing in SimpleITK, without having to go to NumPy.

banikr · August 4, 2021, 4:59pm

This solution worked perfectly:

(array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]),
 array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]))

Also shows same properties in ITKsnap. Didn’t know ITKsnap changes the spacing for visualization.
Thanks!

banikr · August 4, 2021, 10:15pm

This also works perfectly.

>> imgnb.affine, segnb2.affine
(array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]),
 array([[  -0.36132801,    0.        ,    0.        ,   69.19999695],
        [   0.        ,   -0.36132801,    0.        ,  109.80000305],
        [   0.        ,    0.        ,    0.625     , -240.375     ],
        [   0.        ,    0.        ,    0.        ,    1.        ]]))

Thanks!

banikr · August 7, 2021, 9:18pm

Hello @dzenanz
When I try to write the Binary SegITK image as nifti file, I get the following error:

sitk.WriteImage(SegItk, nrd2niiPath)
  File "/home/banikr2/miniconda3/envs/carotid36/lib/python3.6/site-packages/SimpleITK/extra.py", line 394, in WriteImage
    return writer.Execute(image)
  File "/home/banikr2/miniconda3/envs/carotid36/lib/python3.6/site-packages/SimpleITK/SimpleITK.py", line 5518, in Execute
    return _SimpleITK.ImageFileWriter_Execute(self, *args)
TypeError: Wrong number or type of arguments for overloaded function 'ImageFileWriter_Execute'.
  Possible C/C++ prototypes are:
    itk::simple::ImageFileWriter::Execute(itk::simple::Image const &)
    itk::simple::ImageFileWriter::Execute(itk::simple::Image const &,std::string const &,bool,int)

Any idea how to solve it?

MRMR · March 7, 2022, 3:52pm

Hi @banikr,

I am struggling with the same point, can you share with me how did you managed to convert from dicom/nrrd to nifty keeping the origins matched?