Reorienting Images

Algorithms
,
1

Hello, I am trying to achieve the attached process (which I had been doing using nibabel) with SimpleITK. I’m not sure how to interpret the cosine matrix, though, so I’m unclear about how to ensure L-P-S orientation:

DESIRED_ORIENTATION = ('L', 'P', 'S')

def extract_orientation(ctx):
    ctx.affine = ctx.mos_img.affine
    ctx.orientation = nib.aff2axcodes(ctx.affine)


def reorient_mask(ctx, desired_orientation):
    ctx.orig_orientation = ctx.orientation

    # Convert the orientations to the orientation matrices
    current_ornt = nib.orientations.axcodes2ornt(ctx.orientation)
    desired_ornt = nib.orientations.axcodes2ornt(desired_orientation)
    
    # Get the transform from current to desired orientation
    ctx.transform_ornt = nib.orientations.ornt_transform(current_ornt, desired_ornt)
    
    # Apply the transform
    ctx.orig_mos_data = ctx.mos_data
    ctx.mos_data = nib.apply_orientation(ctx.mos_data, ctx.transform_ornt)

    # Update the affine matrix
    ctx.orig_affine = ctx.affine
    ctx.affine = ctx.affine @ nib.orientations.inv_ornt_aff(ctx.transform_ornt, ctx.mos_img.shape)

    # Reorient the image
    ctx.orig_img = ctx.mos_img
    ctx.mos_img = nib.Nifti1Image(ctx.mos_data, ctx.affine)

    # Update the orientation in the context object
    extract_orientation(ctx)
2

Hello @dlevy-ph,

The class you are looking for is called DICOMOrientImageFilter, so this should work for you sitk.DICOMOrient(image, "LPS").

3

To clarify, with the DICOMOrientImageFilter the notation is the “To” direction for the label of axes. That is LPS means { right-to-Left, anterior-to-Posterior, inferior-to-Superior}.

1 Like