Pulmonary lobe segmentation from CT images using fissureness, airways, vessels and multilevel B-splines

Lobe detection from CT images is a challenging segmentation problem with important respiratory health care applications, including surgical planning and regional image analysis. We present a fully automated method for segmenting the pulmonary lobes. We first build a lobar approximation by applying a watershed transform to a vesselness density filter, using seed points generated from segmentation and analysis of the bronchial tree. We then apply a fissureness filter, which combines Hessian-based detection of planar structures with suppression of locally fissure-like points on the boundaries of the pulmonary vasculature. Finally, we fit a smooth multi-level B-spline curve through the fissureness maxima and extrapolate to the lung boundaries. Our method addresses several limitations of similar work, namely it is robust to incomplete fissures and vessels crossing the lobar boundaries, and it is computationally efficient and does not require training. We provide validation using fissure landmarks manually placed on 10 lung cancer datasets by a pulmonary clinician.