An oriented derivative of stick filter and post-processing segmentation algorithms for pulmonary fissure detection in CT images

Abstract

Abstract Knowledge of pulmonary fissure anatomy is valuable in localization of lesions and evaluation of lung disease. Under CT imaging, pulmonary fissure detection is an intricate task due to factors such as pathological deformation, partial volume effect and intensity variability. To solve the problem, an oriented derivative of stick (ODoS) filter and a post-processing segmentation algorithm are introduced for pulmonary fissure detection. Here, the ODoS filter is proposed as an improvement to an existing derivative of stick (DoS) filter by merging the stick orientation information for pulmonary fissure enhancement, which will increase its clutter discriminating ability especially on those linking to the fissure object. Based on an observation that the pulmonary fissures often appear as coplanar structures and have similar directions across the sagittal plane, we present an orientation partition scheme for fissure patches and noise separation in different orientation partition. After removing the small-sized noise, the purified patches from different partitions are iteratively integrated by a fissure patches integration scheme for pulmonary fissure segmentation. With the additional direction constraint, the ODoS filtering response can be more completely detected and the noise residual could be kept at the lowest level. The performance of our algorithms is validated in experiments with a publicly available challenge dataset, i.e. the LObe and Lung Analysis 2011 (LOLA11) data, including 55 CT scans. Compared with manual references, the proposed method acquired a high median F1-score of 0.877. The effectiveness of our scheme was verified by visual inspection and quantitative evaluation.