Automated integer programming based separation of arteries and veins from thoracic CT images

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Pulmonary Vascular Analysis

Background:

• Accurate analysis of lung vessels is crucial for non-invasive diagnosis of lung diseases.
• Distinguishing between pulmonary arteries and veins is essential for detecting specific vascular changes.

Purpose of the Study:

• To present a novel, fully automatic method for separating arteries and veins in thoracic computed tomography (CT) images.
• To combine local and global vessel properties for improved artery-vein separation.

Main Methods:

• Vessel subtree extraction using an integer program (IP) reformulated as a Markov Random Field.
• Artery-vein labeling using a second IP based on anatomical properties like distribution and proximity to bronchi.
• Evaluation against 25 manual reference segmentations.

Main Results:

• Subtree extraction demonstrated minimal non-vascular structures (median 0.9%) and merged subtrees (median 0.6%).
• The artery-vein separation achieved a high median voxel-based overlap of 96.3% with manual segmentations.
• The method outperformed a state-of-the-art interactive approach.

Conclusions:

• The novel automated method effectively separates pulmonary arteries and veins in CT images.
• This technique shows potential as an integral part of computer-aided pulmonary diagnosis systems.
• Accurate artery-vein separation enhances the diagnostic capabilities for lung diseases.