Quantitative measurement of vasculature in coronary angiogram videos

Current diagnosis of coronary artery disease relies on visual examination of angiograms by operators to identify significant stenoses in arteries. The significant limitations of this approach are both under and over-calling of stenoses. Although techniques such as quantitative coronary angiography (QCA) are available, these still require human input and resource costs. Thus, an automated system is required to quantitatively analyze angiograms. Although there are some methods for segmenting the vessels, these methods are mostly based on artifacts-free assumption. In this paper we propose a fully-automated system for analyzing X-ray coronary angiograms in which an angiogram video is analyzed first to select the frames containing maximum visible vasculature. Then, an accurate vascular network skeleton is generated and a profile of vessel thickness along the skeleton is estimated. Moreover, the proposed method calculates the tortuosity to remove the catheter from the images. The accuracy of the proposed method is assessed with ground truth generated manually by cardiologists.