Automatic Coronary Artery Segmentation in X-ray Angiograms by Multiple Convolutional Neural Networks

Accurate coronary artery segmentation in X-ray angiographic images is a challenging task due to the low image quality and presence of artifacts. This paper proposes an automatic vessel segmentation method in the X-ray angiographic images using correspondence matching and convolutional neural networks (CNN). First, a dense correspondence between the live image and the mask image is generated. Second, patches from live images as well as patches from mask images are put into a two-channel network to achieve a coarse segmentation for the region of interest. Third, a one-channel CNN is used to generate the fine segmentation result. Experiments demonstrate that our method is very effective and robust for coronary artery segmentation, which is better than the other three state-of-the-art methods.