Three-Step Segmentation of the Lower Airways with Advanced Leakage-Control

Segmentation of the tracheo-bronchial tree of the lung serves as an important tool for diagnosis and treatment planning of various pathologies, e.g. by allowing accurate volume measurements or detecting malformations. However, segmenting the airways of the human lung is notoriously difficult. This is due to the small size of the participating anatomical structures, which are moreover subject to partial volume and noise effects. Limited intensity contrasts between air and lung parenchyma also complicate segmentation. In this paper, we present our hybrid segmentation method consisting of three main steps, which are iterated until a satisfactory result is achieved. User interaction is limited to the specification of a seed point inside the easily detectable trachea. Further, we discuss the performance of our method within the EXACT09 challenge, where 20 test datasets with varying quality and pathologies had to be processed.