Segmentation of 3D Pulmonary Trees Using Mathematical Morphology

Abstract

We propose algorithms to automate the segmentation of pulmonary tree structures in the lung, using tools from Mathematical Morphology. This involves segmenting three different types of three-dimensional tree structures (airway tree, pulmonary artery, pulmonary vein) from a stack of grayscale Computed Tomography (CT) images. The proposed algorithms rely on the grayscale reconstruction operator to extract potential tree regions in each of the CT images. A three-dimensional seeded region growing is performed on the processed stack of images to obtain the segmented tree volumes. We first segment the airway tree, and use the geometric features (shape & size) of its segmented output to segment the pulmonary arterial and veinous trees. Segmentation results of pulmonary tree volumes obtained from CT image stacks of a static dog lung are very encouraging and we intend to apply these techniques on dynamic lung data in a clinical setting.