This repository provides Python scripts for performing initial coarse segmentation of pulmonary airways and blood vessels from computed tomography (CT) images. These tools implement the automated pre-segmentation pipeline described in the [Link to your Data Descriptor paper/preprint, if available, otherwise mention it's associated with a manuscript] for generating the AirRC dataset.

The primary goal of these scripts is to produce rough initial segmentations that can significantly expedite the manual annotation process by providing a starting point for expert radiologists. The final accuracy of segmentations for datasets like AirRC relies on subsequent meticulous manual refinement.

• Airway Segmentation:
  • Seed-based 3D region growing.
  • Two methods provided:
    1. ConfidenceConnectedImageFilter from SimpleITK.
    2. ConnectedThresholdImageFilter from SimpleITK, applied on HU-clamped images, followed by connected component volume filtering to remove leakage into lung parenchyma.
  • Preprocessing includes HU value clamping specific for airway visualization.
  • Optional morphological refinement (opening and closing).
• Pulmonary Vessel Segmentation:
  • 2D slice-wise processing.
  • Lung field approximation using Otsu's thresholding followed by refinement steps (border clearing, hole filling, morphological smoothing).
  • Adaptive thresholding for vessel candidate identification based on non-lung pixel intensity, after contrast enhancement.
  • Outputs a 3D binary mask of potential vessel structures.

• pulmonary_structure_segmentation_tools.py: Contains all core functions for reading images, performing airway segmentation, performing vessel segmentation, and writing output masks. Includes an example usage block (if __name__ == "__main__":) for demonstration.

• Python 3.8+
• Required Python packages:
  • SimpleITK
  • NumPy
  • OpenCV-Python (opencv-python)
  • Scikit-image (scikit-image)

You can install the required packages using pip:

pip install SimpleITK numpy opencv-python scikit-image