NIfTI to RTStruct - no registration with CT (rotation problem?)

[balduz94]

Hello, I am having problems with this code: I have many structures in NIfTI format that, once I create the RTStruct file, do not correctly overlap with the original CT. Even applying rotations as suggested by other users, I cannot achieve a good overlap with the CT. However, if I open these structures directly with 3D Slicer, they overlap perfectly, so it is not a segmentation issue. I am leaving the code here hoping for some help. Thank you, Emanuele.

from rt_utils import RTStructBuilder
import nibabel as nib
import numpy as np
import os
import colorsys

def generate_unique_colors(n):
    colors = []
    for i in range(n):
        hue = i / n
        saturation = 0.7  # Scegli un valore di saturazione medio-alto
        value = 0.9  # Scegli un valore di luminosità medio-alto
        rgb = colorsys.hsv_to_rgb(hue, saturation, value)
        colors.append([int(x * 255 + 0.5) for x in rgb])
    return colors

folder_path = r"\path_folder_with_nifti_segmentation"
dicom_series_path = r"\path_CT_reference"

rtstruct = RTStructBuilder.create_new(dicom_series_path)

nii_files = [f for f in os.listdir(folder_path) if f.endswith('.nii.gz')]
unique_colors = generate_unique_colors(len(nii_files))

for file, color in zip(nii_files, unique_colors):
    nii_path = os.path.join(folder_path, file)
    nii_image = nib.load(nii_path)
    data = nii_image.get_fdata() > 0

    mask_dataflip = np.flipud(data)
    mask_dataflip2 = np.flip(mask_dataflip)
    mask_rtdata = np.rot90(mask_dataflip2)

    roi_name = os.path.splitext(file)[0]

    rtstruct.add_roi(mask=mask_rtdata, color=color, name=roi_name)

rtstruct.save(r'\path_RTStruct_output\new-rt-struct.dcm')

[redvens]

Have you solved this problem?

[winterkid2000]

I had a same problem too, it was a matter of transpose from xyz to yxz. For rt utils, dicom files are automatically rotated from xyz to yxz but not the nii mask file(mask arrray I mean). So, you should try

nii_path = os.path.join(folder_path, file)
nii_image = nib.load(nii_path)
data = nii_image.get_fdata() > 0
data = np.transpose(data, (1, 0, 2))

In medical image -especially DICOM Numpy-, the array of axis are set like (Y, X, Z) and because Y stands for row(AP), X stands for column(LR).

[winterkid2000]

additionally, because roi mask is just numpy array so they can't recognise it with meta-data like spacing, orientation or something so we are required to transpose it.