Purpose: Because vasculature is involved in many diseases, vasculature segmentation is an important middle-level image processing task for supporting higher-level tasks such as lung ventilation function measurement, motion estimation, treatment response evaluation and disease diagnosis. In this study, we developed a comprehensive procedure to segment high-resolution vasculature tree in CTs, and to detect vessel bifurcations, i.e., the anatomically stable landmarks. The success of the proposed work will support future studies to automatically detect stable landmark pairs between pair of CTs and to provide high-quality high-density benchmark datasets to verify deformable image registration algorithms.
Methods: A novel workflow was developed to segment high-resolution vasculature trees in CTs. Important steps are 1) segmenting target organ, 2) denoising image to enhance the contrast between the vessel and the background, 3) computing vesselness, 4) hysteresis thresholding the computed vesselness map to automatically segment the vessel tree on the vesselness map, 5) pruning the segmented vessel tree, 6) thinning the vessel tree to detect the skeleton, and 7) detecting the vessel bifurcation points on the skeleton tree.
Results: The workflow was successfully applied on lung and liver CTs. In lung CTs with ≤1 mm slice thickness, terminal bronchioles after 9 levels of bifurcations were precisely segmented, and vessel tree bifurcations are precisely detected. The number of detected stable landmarks was ~1500 per lung. For liver CTs with much lower vessel-to-background SNR, the number of detected stable landmarks was 80 to 100.
Conclusion: The proposed workflow can segment high-resolution vasculature trees in CTs for lung and liver and detect stable landmark points at the vessel bifurcations. It can be used with the published landmark pairing methods to produce large amount of stable landmark pairs, that will be used to quantitatively evaluate the image registration methods and results.