Airway tree segmentation using adaptive regions of interest

Abstract

The accurate segmentation of the human airway tree from volumetric CT images builds an important corner stone in pulmonary image processing. It is the basis for many consecutive processing steps like branch-point labeling and matching, virtual bronchoscopy, and more. Previously reported airway tree segmentation methods often suffer from "leaking" into the surrounding lung tissue, caused by the anatomically thin airway wall combined with the occurrence of partial volume effect and noise. Another common problem with previously proposed airway segmentation algorithms is their difficulties with segmenting low dose scans and scans of heavily diseased lungs. We present a new airway tree segmentation method that works in 3D, avoids leaks, and automatically adapts to different types of scans without the need for the user to iteratively adjust any parameters.