3D segmentation of the tracheobronchial tree using multiscale morphology enhancement filter

Abstract

In this article we present a new region growing algorithm for airway segmentation based on multiscale black tophat enhancement filter. Lung airways are tubular structures that display specific characteristics, such as highly variable intensity levels within the lumen and proximity to vessels. The proposed airways enhancement filter aims to separate airways from adjacent lung parenchyma and vessel. Based on the filter ouput, the region growing is performed in order to delineate the airways and then to reconstruct the tracheobronchial tree. The proposed method has been applied on various CT scans. In this paper, an experimental comparison study between our filter and the "gold standard" filters used to enhance tubular structures (Frangi, Sato and Krissian filters) followed by a region growing process is performed on data from the VESSEL12 challenge framework. Our approach outperforms the other considered methods in terms of retrieved bronchi and computing time.