How to overwrite spacing of ScalarImage ?

[Optimox]

Hi,

I have several 3D stacks and annotations that can be either in dicom or tiff format.

The spacing are not necessarily correct inside the dicom or tiff files so when I try to access data with augmentation from the dataloader I end up with this error:

RuntimeError: More than one value for "spacing" found in subject images:
{'image': (0.001002435076546, 0.001, 0.0010000000474974491),
 'mask': (1.0000000150184933, 1.0000000150184933, 1.0)}

But I don't want to Resample my data as I know that the scaling are wrong and that one voxel in one image correspond to one voxel in one segmenttation.

So I tried to overwrite the spacing of my images:

my_img = tio.ScalarImage(img_path)
my_img.spacing = (1.0, 1.0, 1.0)

But then I get the bellow error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/tmp/ipykernel_2544/1565851587.py in <module>
----> 1 my_img.spacing = (1.0, 1.0, 1.0)

AttributeError: can't set attribute

So here is my question: how can I overwrite the spacing on a ScalarImage and LabelMap ?

Thanks for the help!

[romainVala]

Hi
not sure how to explicitely set the affine (that will define the spacing). may be @fepegar has an idea how to achieve it ... ?

In the mean time you can try to copy the affine from your image to you label with the transform CopyAffine
Thus you will have 2 coherent affine (and then spacing) but depending on the transform you used after this may not be enough (because having the correct affine is important, for instance for Resampling )

[Optimox]

Thanks for your answer, I am a bit worried about unknown side effects...

[romainVala]

Actually you can trust dicom information, but I am not sure on the tiff one (this is not a volume format)
So copy the affine from the image (dicom) to replace the one from the label (in tiff, I guess)

[fepegar]

TIFF images can contain a stack. I agree that CopyAffine is probably the simplest solution here.