Fully automated segmentation of coronary lumen based on the directional minimal path and image fusion

Abstract

The segmentation of coronary lumen is a challenging but important task in clinical application of cardiac computed tomography (CTA). In this paper, a new method is proposed to segment the coronary lumen in a fully automatic manner. This method is based on the directional minimal path and the level-set segmentation in the 2D fused image. The directional minimal path is first used automatically to track the coronary centerlines of the main branches, which provides the center location of the coronary lumen. Then, based on the coronary centerline, the cross-sectional planes are calculated in the 3D CTA images. In order to increase the successful rate of the lumen segmentation, the gray-filtered and vesselness-enhanced images are calculated respectively in the cross-sectional planes and the 3D stacking of the cross-sectional planes. And, the two enhanced images are fused to generate the fused image. Finally, the level-set algorithm is used to segment the coronary lumen in the cross-sectional planes of the fused image. The proposed method is validated by segmenting the lumen of the three main coronary branches. The DICE (Dice coefficients) are 83.2% (RCA), 81.7% (LAD) and 83.5% (LCX), respectively.