Enhanced Subtraction Image Guided Convolutional Neural Network for Coronary Artery Segmentation

Jingfan Fan, Chenbin Du, Shuang Song, Weijian Cong, Aimin Hao, Jian Yang

Abstract: Digital subtraction angiography (DSA) is a fluoroscopic technique used to clearly visualize blood vessels. However, accurate segmentation of coronary arteries cannot be directly obtained from DSA images because of motion artifacts. In this paper, a fully convolutional network is designed to segment the coronary arteries from DSA images instead of angiographic images. First, an ORPCA method with intra-frame and inter-frame constraints is introduced to enhance the vessel structure in DSA. Then, an enhanced DSA image-guided segmentation network, which is a fully convolutional network composed of an encoder path and a decoder path, is proposed to extract the coronary arteries to learn the vascular features from the enhanced vascular structures. The experimental results demonstrate that the proposed method is more effective and accurate in coronary artery segmentation, compared with state-of-the-art methods.