Automatic Segmentation of Vertebral Arteries in Head and Neck CT Angiography Images

Computed tomography (CT) angiography is a useful imaging technique for the evaluation of the intraand extra-cranial vasculature. With the introduction of multi-detector row CT scanners, head and neck CT angiography images make it easy to detect cerebral aneurysms, arterial stenosis, and other vascular anomalies. To get an arterial anatomy in three-dimensional views, CT angiography images are often displayed using conventional volume rendering and maximum intensity projection (1-3). However, there is the limitation to get a clear arterial anatomy where arteries are contiguous with the bone. Two main approaches are generally used for vessel segmentation in CT angiography. First approach is to subtract enhanced images which taken after injecting a contrast material from nonenhanced images. Gratama et al. proposed a multi-scale matched mask bone elimination using combined high resolution and low resolution image of the non-enhanced scan (4). Kwon et al. proposed a 3D registration to align a CT to CTA using normalized mutual information and subtraction-and-refinement to effectively remove unwanted residuals (5). Hong et al. proposed 3D rigid registration technique based on locally weighted 3D distance map of feature point and bone masking (6). Kim et al. extracted vessels using multi-segmented volume subtraction of lower extremities based on the 3D rigid registration (7, 8). These methods have been shown to be good quality results. However, bone masking requires the acquisition of an additional CT scan, thereby increasing the radiation dose to the patient. Second approach is directly vessel extracting method from CT angiography by using vessel tracking procedure. Shim et al. proposed a vessel segmentation method via the ellipse travelling and tracking in threedimensional space (9). Manniesing et al. proposed a 3D vessel axis tracking based on evolving surfaces using skeleton and shape constraint topology (10). Gulsun at al. proposed a centerline tracking based on multi-scale medialness filters and lumen boundary segmentation with graph-cut optimization (11). These methods typically require user interaction in the form of one or more seed points placed in the vessel to be extracted. In this paper, we propose a fully automated vessel segmentation method of vertebral arteries in head and neck CT angiograOriginal Article