Preprocess DICOM and RTSTRUCT

Load a CT DICOM series and its RTSTRUCT GTV-1 ROI. Resample the image and mask to 2 mm in-plane spacing while retaining the original slice spacing. Use the matching series directory, RTSTRUCT path, and structure name for your data; see Expected data structure and Resampling guidelines.

from zrad.image import Image
from zrad.preprocessing import ImageResampler, MaskResampler, RoiData

# The RTSTRUCT mask is converted onto the CT image grid.
image = Image.from_dicom("path/to/dicom_series", modality="CT")
mask = Image.from_dicom_mask(
    "path/to/rtstruct.dcm", "GTV-1", reference=image
)
roi = RoiData(image=image, morphological_mask=mask)

# Keep the original through-plane spacing for slice-wise resampling.
resolution = (2.0, 2.0, float(image.spacing[2]))
roi = ImageResampler(resolution, method="linear").apply(roi)
roi = MaskResampler(
    resolution, method="linear", partial_volume_threshold=0.5
).apply(roi)

assert tuple(roi.image.shape) == tuple(roi.morphological_mask.shape)
assert tuple(roi.image.spacing) == tuple(roi.morphological_mask.spacing)
roi.image.save_as_nifti("output/image.nii.gz")
roi.morphological_mask.save_as_nifti("output/GTV-1.nii.gz")
print(f"Spacing (x, y, z): {tuple(roi.image.spacing)} mm")

The exported CT and mask share a voxel grid. With the bundled phantom, the reported spacing is (2.0, 2.0, 3.0) mm.