combine_segmentations.py

import os
import SimpleITK as sitk
import numpy as np

from glob import glob
from tqdm import tqdm


def combine_segmentations(folder, output_filename="segmentation.nii.gz"):
    """
    Combines multiple single-label segmentation files into a single multi-label segmentation file.

    Args:
        folder (str): Path to the folder containing segmentation files.
        output_filename (str): Name of the combined multi-label segmentation file.
    """
    # Define the segmentation file names and their corresponding labels
    segmentation_labels = {
        "seg-Esophagus.nii.gz": 1,
        "seg-GTV-1.nii.gz": 2,
        "seg-Heart.nii.gz": 3,
        "seg-Lung-Left.nii.gz": 4,
        "seg-Lung-Right.nii.gz": 5,
        "seg-Spinal-Cord.nii.gz": 6,
    }

    # Initialize an empty image for combining segmentations
    combined_image = None

    for seg_file, label in segmentation_labels.items():
        seg_path = os.path.join(folder, seg_file)

        if os.path.exists(seg_path):
            # Read the segmentation file
            seg_image = sitk.ReadImage(seg_path)

            # Convert to numpy array
            seg_array = sitk.GetArrayFromImage(seg_image)

            # Create a binary mask for the current label
            binary_mask = (seg_array > 0).astype(np.uint8) * label

            if combined_image is None:
                # Initialize the combined image with the same size and spacing as the first segmentation
                combined_array = np.zeros_like(seg_array, dtype=np.uint8)
                combined_image = seg_image

            # Add the current binary mask to the combined array (ensuring no label overlap)
            combined_array = np.maximum(combined_array, binary_mask)

    if combined_image is not None:
        # Set the combined array as the new image's data
        combined_image = sitk.GetImageFromArray(combined_array)
        combined_image.CopyInformation(seg_image)

        # Save the combined multi-label segmentation file
        output_path = os.path.join(folder, output_filename)
        sitk.WriteImage(combined_image, output_path)

        print(f"Combined multi-label segmentation saved at: {output_path}")
    else:
        print("No segmentation files found to combine.")


folders = sorted(glob(f'NSCLC-Radiomics-NIFTI/*'))

for fd in tqdm(folders):
    combine_segmentations(fd)