Convert from 3D numpy array to simple ITK

I have a NumPy array (image_array) with the size of 400 * 100 * 800 (spacing: [1, 4, 0.1]).

I used the below code to convert it to sitk image for registration:

image = sitk.GetImageFromArray(image_array)
image.SetOrigin((0, 0, 0))

When I read the spacing from sitk image using image.GetSpacing() it is ok but the size (image.GetSize()) is different: 800 * 100 * 400

Do I need to reshape the NumPy array before importing it to the SITK and then just use the sitk.GetArrayFromImage to convert it to the NumPy array and have the same Spacing and Size?

I tried to inspect the image inSITK using FIJI but it didn’t show the spacing correctly (all are equal to one)

Thank you

Hello @Moha_data,

Welcome to SimpleITK!

SimpleITK and numpy use a different indexing order. SimpleITK uses x-y-z. Numpy uses z-y-x. The GetImageFromArray method accounts for this. Thus when you inspect the image size afterwards the reported order is x-y-z which matches your original numpy size. With respect to SetSpacing, it should also follow the x-y-z order. For additional details see this jupyter notebook.

Fiji is not aware of variable image spacing, assumes isotropic pixels/voxels ([1,1,1]). Other programs that do account for image spacing can be used via the Show method (e.g. ITK-SNAP, 3D Slicer). See this jupyter notebook for details.

1 Like

Thanks @zivy for the comment.
So, assuming we have an Image_array with corresponding spacing, we can use the below code to convert it to sitk image:

image_sitk = sitk.GetImageFromArray(image_array)
image_sitk.SetOrigin((0, 0, 0))

and after any modification in sitk, we can use:

modified_image = sitk.GetArrayFromImage(modified_image_sitk)`

and spacing.reverse() to reorder the spacing to be used in NumPy.

Hello @Moha_data,

Yes (assuming spacing is a list associated with the numpy data so spacing.reverse() changes it in place).

1 Like