Segmentation of obstructed airway branches in CT using airway topology and statistical shape analysis

Chest pathology can lead to airway branch obstruction, and segmentation of the airways beyond obstructions is a challenge. We propose a novel method that automatically identifies points of obstruction using airway topology and statistical shape analysis and segments disconnected branches. The point of obstruction is used to define an allowed region for the airway beyond the obstruction in order to direct the segmentation. This method can be used to extend standard airway segmentation approaches and was evaluated using 42 chest CT scans of paediatric patients with tuberculosis. The algorithm was compared to manually labelled obstructions, and identified 24 of the 26 obstructed branches (where the 2 missing branches would likely be identified with more training data for the left main bronchus) and identified 18 of the 19 disconnected airway regions.