科研之友
微信
新浪微博
Facebook
分享链接
摘要
Coronary artery disease (CAD) is a leading cause of death worldwide. Although coronary CT angiography (CTA) and other new technologies emerge increasingly, conventional coronary angiography (CCA) remains as the gold standard for diagnosis of CAD, and the only way to be involved in the interventional surgery. Centerline extraction of the coronary arteries is the essential information for radiologists, and is also the foundation for a computer-aided detection (CADe) system to assist them. As the data is obtained more and more, manual extraction is impractical, a fully automatic extraction method is necessary for radiologists. However, due to the projection nature, the extraction of vessels becomes extremely difficult because of non-uniform stating caused by the contrast agent distribution and overlap of the organs. Furthermore, the shape of the blood vessels is another important information needed in clinical practice, but their identification is challenging, especially at the intersectional positions. In this paper, we propose a method to extract the blood vessel contour and identify their shapes at the intersections simultaneously. Firstly, we refine Frangi's detection result to compensate the vesselness measure, ensure connectivity and eliminate artifacts as far as possible. Secondly, we study a vessel connectedness based clustering method to identify the each blood vessel. Thirdly, in order to handle the gaps and holes in enhanced vessel image, we employ a robust method based on principle curves to extract the centerlines. Finally, We evaluate the performance of our method on 60 clinical samples in angiographies. The method performs well with respect to centerline extraction, which its average accuracy is 96.247%, sensitivity is 79.981% and specificity is 97.754%.
