Search

Positron emission tomography (PET) combined with computed tomography (CT) is a routine imaging modality for the diagnosis and interpretation of malignant diseases of the thorax. Accurate airway segmentation is critical for the localization of sites of abnormal metabolism detected with PET-CT. The vast majority of published segmentation algorithms, however, are designed for high- resolution CT and these algorithms do not perform well with the low-contrast CT acquired in PET-CT images. In this study, we present a new fully automated airway segmentation algorithm that is optimised to tolerate the image characteristics inherent in low-contrast CT images. Our algorithm accurately and robustly segments the airway by introducing: (i) a robust multi-atlas initialisation which incorporates shape priori knowledge for seeds derivation; and (ii) a modified knowledge-based random walk segmentation that uses the derived seeds and manipulates the weights of the edge paths in a locally constrained search space. Our proposed algorithm was evaluated on 20 clinical low-contrast CT from PET-CT patient studies and demonstrated better performance in segmentation results against comparative state-of- the-art algorithms.