Getting affine matrix error using loadimaged to read nii files.

[Minxiangliu]

I want to use Monai to read the nii mask image from Pmod output. The original image can be read normally, but there is an error when reading the mask image. How should I solve it?
Thanks in advance!

monai version: 0.8.1
File Link : NPC_AI_Detection_01 _SUVconversion.nii
File Link : NPC_AI_Detection_01_2.5_TN_merge.nii

from monai.transforms import Compose, LoadImaged
def toLoadNiiFile(path):
     trans = Compose([ LoadImaged(keys=['img']), EnsureTyped(keys=['img'])])
     return trans({'img':path})

toLoadNiiFile('NPC_AI_Detection_01 _SUVconversion.nii') # Is normally.
toLoadNiiFile('NPC_AI_Detection_01_2.5_TN_merge.nii')   # Is error

with open NPC_AI_Detection_01_2.5_TN_merge.nii

Modifying image pixdim from [3.13295 3.13295 2.027   1.     ] to [  3.13295007   3.13295007   2.02699995 611.41126719]
Modifying image pixdim from [4.6875 4.6875 3.27   1.    ] to [  4.6875       4.6875       3.26999998 593.6261095 ]
Modifying image pixdim from [3.13295 3.13295 2.027   1.     ] to [  3.13295007   3.13295007   2.02699995 656.51590521]
Modifying image pixdim from [4.6875    4.6875    3.2698658 1.       ] to [  4.6875       4.6875       3.26986575 643.87348714]
Modifying image pixdim from [3.13295 3.13295 2.027   1.     ] to [  3.13295007   3.13295007   2.02699995 704.34331593]
Modifying image pixdim from [3.13295   3.13295   2.0269978 1.       ] to [  3.13295007   3.13295007   2.0269978  656.51552361]
Modifying image pixdim from [3.13295   3.13295   2.0270104 1.       ] to [  3.13295007   3.13295007   2.02701044 656.51777084]
Modifying image pixdim from [5.46875   5.46875   3.2702014 1.       ] to [  5.46875      5.46875      3.27020144 691.81655883]
Modifying image pixdim from [3.13295   3.13295   2.0269938 1.       ] to [  3.13295007   3.13295007   2.02699375 656.5148028 ]
Modifying image pixdim from [3.13295 3.13295 2.027   1.     ] to [  3.13295007   3.13295007   2.02699995 605.12678054]
Modifying image pixdim from [3.13295  3.13295  2.026996 1.      ] to [  3.13295007   3.13295007   2.0269959  656.5151844 ]
Modifying image pixdim from [3.13295   3.13295   2.0270073 1.       ] to [  3.13295007   3.13295007   2.02700734 704.34487804]
Modifying image pixdim from [5.46875   5.46875   3.2699237 1.       ] to [  5.46875      5.46875      3.26992369 651.67485818]
Traceback (most recent call last):
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\transforms\transform.py", line 89, in apply_transform
    return _apply_transform(transform, data, unpack_items)
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\transforms\transform.py", line 53, in _apply_transform
    return transform(parameters)
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\transforms\io\dictionary.py", line 121, in __call__
    data = self._loader(d[key], reader)
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\transforms\io\array.py", line 216, in __call__
    img_array, meta_data = reader.get_data(img)
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\data\image_reader.py", line 454, in get_data
    _copy_compatible_dict(header, compatible_meta)
  File "D:\Anaconda3\envs\monai\lib\site-packages\monai\data\image_reader.py", line 120, in _copy_compatible_dict
    "affine matrix of all images should be the same for channel-wise concatenation. "
RuntimeError: affine matrix of all images should be the same for channel-wise concatenation. Got [[  -4.6875        0.            0.          297.65625   ]
 [   0.            4.6875        0.         -297.65625   ]
 [   0.            0.            3.26999998 -418.55999756]
 [   0.            0.            0.            1.        ]] and [[  -3.13295007    0.            0.          311.72853172]
 [   0.            3.13295007    0.         -311.72853172]
 [   0.            0.            2.02699995 -423.64298964]
 [   0.            0.            0.            1.        ]].

[wyli]

the meta info doesn't match, so it doesn't make sense to stack them into a single array.
if you still want to do that, it's possible to set track_meta=False when loading and use monai.transforms.ConcatItemsd:

from monai.transforms import Compose, EnsureTyped, LoadImaged, Orientationd, EnsureChannelFirstd, ConcatItemsd


def toLoadNiiFile(path):
     trans = Compose([ LoadImaged(keys=['img']), EnsureChannelFirstd(keys="img"), Orientationd('img', "RAS"), EnsureTyped(keys=['img'], track_meta=False)])
     return trans({'img':path})['img']

t1 = toLoadNiiFile('NPC_AI_Detection_01 _SUVconversion.nii') # Is normally.
t2 = toLoadNiiFile('NPC_AI_Detection_01_2.5_TN_merge.nii')   # Is error
print(t1.shape)
print(t2.shape)
output = ConcatItemsd(keys=['img1', 'img2'], name='img3')({'img1': t1, 'img2': t2})
print(output['img3'].shape)

( this script requires monai 0.9.1)

[Minxiangliu]

Hi @wyli ,
Thank you very much for your reply, I want to ask another question, my purpose here is not to concat the two data, I want to read the two data through monai, select the slice range and output nii. gz file, suppose I changed the number of slices of the original data, so the meta information needs to be changed and saved?
Like as:

from monai.transforms import Compose, EnsureTyped, LoadImaged, Orientationd, EnsureChannelFirstd, SaveImage
import numpy as np

def toLoadNiiFile(path):
     trans = Compose([ LoadImaged(keys=['img'], meta_keys='meta'), EnsureChannelFirstd(keys="img"), Orientationd('img', "RAS"), EnsureTyped(keys=['img'], track_meta=False)])
     return trans({'img':path})

data = toLoadNiiFile('NPC_AI_Detection_01 _SUVconversion.nii')
img, meta = data['img'], data['meta']
img = img[0,:,:,30:80]
meta['spatial_shape'] = np.array(img.shape, dtype=np.int16)
saver = SaveImage(output_dir='new', mode='bilinear', output_postfix='')
saver(img, meta_data=meta)

[wyli]

it's possible in v0.9.1 and newer versions with the metatensor, and SpatialCropd does the slicing:

from monai.transforms import Compose, EnsureTyped, LoadImaged, Orientationd, EnsureChannelFirstd, SaveImaged, SpatialCropd
import numpy as np

def toLoadNiiFile(path):
     trans = Compose([ LoadImaged(keys=['img'], meta_keys='meta'), EnsureChannelFirstd(keys="img"), Orientationd('img', "RAS"), EnsureTyped(keys=['img'])])
     return trans({'img':path})

data = toLoadNiiFile('NPC_AI_Detection_01 _SUVconversion.nii')
data = SpatialCropd(keys="img", roi_start=30, roi_end=80)(data)
saver = SaveImaged(keys="img", output_dir='new', output_postfix='', resample=False)
saver(data)