Automatic segmentation of coronary lumen based on minimum path and image fusion from cardiac computed tomography images

Abstract

Segmentation of the coronary lumen is a challenging aspect in the clinical application of cardiac computed tomography. In this paper, a new method is proposed for automatic segmentation of the coronary lumen. The proposed method is based on the directional minimum path and level-set segmentation on a cross-sectional fusion image. The directional minimum path is first used to automatically track the centerlines of the main coronary branches along a region of interest, providing the location of the center of the coronary lumen. Then, based on this centerline, cross-sectional images of the cardiac computed tomography volume are calculated to reconstruct a three-dimensional stacked image. To increase the success rate of lumen segmentation, the three-dimensional stacked image was enhanced using the proposed fusion method. The proposed fusion method included a gray filter that decreases noise, and a measure of the Vesselness that can enhance the coronary structure. Adaptive weighted fusion is then proposed to fuse the enhanced images. Finally, a level-set algorithm was used to segment the coronary lumen in the cross-sectional fused images. The proposed method was validated for the three main coronary branches. The Dice coefficients were 83.8% for the right coronary artery, 82.5% for the left anterior descending artery, and 83.9% for the left circumflex artery, respectively.