Pulmonary Blood Vessels and Nodules Segmentation via Vessel Energy Function and Radius-Variable Sphere Model

Abstract

To help diagnose the early stage of lung cancer, this paper studies pulmonary nodule and blood vessel detection and segmentation. Owing to the fact that variation in the shape and number of pulmonary blood vessels would reveal the progress of lung cancer, automatic segmentation of pulmonary nodules and blood vessels is desirable for chest computer-aided diagnosis (CAD) systems. The proposed algorithm is composed of four steps: pre-segmentation, structure enhancement, active evolution, and refinement. Through the initial extraction of 3D region growing, the line structure of vessel and blob-like structure of nodule would be enhanced by multi-scale filtering. In particular, the active evolution is devoted to the maximum likelihood estimation with a vessel energy function (VEF) of intensity, gradient, and structure. The VEF aims to shape a precise extraction by adapting all the cue distribution along the vessel region from nodules. Furthermore, a radius-variable sphere model is adopted to refine the contour with the smoothness of radius alone the centerline of the blood vessel. Finally, the proposed scheme is sufficiently evaluated to exceed the existing techniques on lung image database consortium (LIDC) database and DICOM images.