Automated airway segmentation from chest ct images combined uniform and local intensities and airway topology structure

High-resolution multi-slice computed tomography (CT) receives high prevalence among pulmologists and clinicians due to its allowance of investigating pulmonary and airway function, and detecting and diagnosing pulmonary lesions. In presence of the specific tree-like morphology, and the impact of imaging defects, abnormalities or other interference factors, the academic community has carried a number of efforts on airway segmentation from chest CT images. However, it is still a challenging problem to detect airways as complete as possible without over-segmentation under a low time cost. In this paper, a fully automated airway segmentation method was presented. The method pipeline included four steps: trachea and main bronchi segmentation, a uniform intensity based larger branch segmentation, a local intensity based smaller branch segmentation, and a topology structure based tiny branch segmentation. A region growing model and a leakage control model was proposed for intensity based steps. In both models all the thresholds were determined selfadaptively to ensure the robustness against imaging diversity. The proposed method was evaluated on a public platform to show its performance.