Abstract

Info

Title:: Coronary artery centerline extraction method based on fully convolutional network

Author(s):: CUI Aoyu; XU Xingfen; TIAN Miao; ZHANG Lei; School of Mathematics and Physics, Lanzhou Jiaotong University, Lanzhou 730070, China

Keywords:: Keywords: image processing centerline extraction fully convolutional network coronary angiography deep learning

Abstract:: Abstract: The accurate extraction of coronary artery centerline is of great significance for the diagnosis of vascular diseases. Herein a coronary artery centerline extraction method based on fully convolutional network is proposed. The Frangi filter based on Hessian matrix is used to remove most of the pseudo-vessel noise, highlight the blood vessel tree of coronary artery, and clearly distinguish the blood vessels from the background and the Steger algorithm is used to obtain the sub-pixel points in the center of the blood vessel cross-section. The initially extracted centerlines are taken as a data set, and a total of 473 processed images are labeled, with 378 in training set and 95 in test set. The test results are evaluated in terms of pixel accuracy (PACC), mean pixel accuracy (MACC), and mean intersection over union (MIoU). The FCN model is used to segment the data set, fuse the low-level feature information with high-level information, and perform deconvolution on the fused feature map. The highest PACC, MACC and MIoU reach 0.85, 0.92 and 0.82, respectively. The proposed method which has high accuracy and can effectively extract the coronary artery centerline can serve as an auxiliary method for the diagnosis of coronary heart diseases.