Ling Fu, Kai Zhao, and Yan Kang
segmentation refinement, multi-branches centerline extraction, cerebral and carotid artery, computer-aided diagnosis
Accurate segmentation of cerebral and carotid arteries is paramount for the detection of vascular disease during computer-aided diagnosis. Accuracies of 100%, however, are impossible to achieve for any automatic methods, especially when these methods are applied to cerebral and carotid arteries, due to the torturous nature of these arteries. For this reason, we propose two segmentation refinement methods which can correct erroneous segmentation results and ensure accuracy. For segmentation leaking into bones or veins, we propose a segmentation-volume partition method to remove these wrongly segmented tissues. Since this method requires knowledge of the centerlines of all branches of a vessel tree, we introduce a centerline extraction method first which develops the distance transform method into multi-branches centerline extraction. Moreover, a locally adaptive method is proposed to add missing vessels failing to be segmented due to, for example, low density of the contrast agent. The quantitative evaluation results of the proposed methods on real images prove the accuracy and repeatability of the proposed methods.
Important Links:
Go Back
