Vessel-guided airway tree segmentation: A voxel classification approach

Area of Science:

• Medical Imaging
• Pulmonary Medicine
• Computer-Aided Diagnosis

Background:

• Accurate segmentation of the airway tree in medical images is essential for diagnosing and monitoring lung diseases.
• Previous methods often struggle with incomplete or low-quality segmentations, limiting their clinical utility.

Purpose of the Study:

• To develop and evaluate a novel airway tree segmentation method combining a trained airway appearance model with vessel orientation information.
• To improve the completeness and accuracy of airway segmentation in low-dose computed tomography (LDCT) scans.

Main Methods:

• A voxel classification approach was used to train an airway appearance model, differentiating airway from non-airway voxels.
• A vessel orientation similarity measure was introduced to enhance segmentation in regions with weak appearance model responses.
• Region growing was applied, integrating the appearance model and orientation information.

Main Results:

• The proposed method significantly improved airway segmentation completeness, yielding 20% longer trees and 50% less leakage compared to baseline approaches.
• Combining the airway appearance model with vessel orientation similarity further enhanced results, showing a significant (p<0.01) 7% increase in correctly extracted branch length.

Conclusions:

• The developed airway segmentation method offers superior performance in terms of completeness and accuracy.
• The integration of a trained appearance model and vessel orientation information represents a significant advancement for airway tree segmentation in LDCT imaging.