Are sitk Numpy array indices zxy in physical or voxel orientation?

I need to get axial, coronal and sagital slices for any image.
I thought for some time that the indices of numpy arrays from images are in ijk so i needed to do some reorienting though i just came to the conclusions that the zxy indices of numpy arrays might already be in physical orientation (LPS).
If that’s true, then sagital slicing is along the 3rd dim (x), axial is along 1st dim (z) and coronal is along second dim (y).
Is that it?

In general you are correct as long as the image’s direction cosine matrix is the identity. If not, then the order of the axis may change (happens in MR when using a set of coronal or sagittal planes during acquisition). For example the MR images used in this Jupyter notebook.

If the direction matrix is not identity, it transforms ijk to LPI, so the rows are L P I vectors in terms of ijk; then it is back to the case above, correct?