TY - JOUR
T1 - High-quality coronary artery segmentation via fuzzy logic modeling coupled with dynamic graph convolutional network
AU - Dong, Caixia
AU - Dai, Duwei
AU - Li, Yang
AU - Xu, Songhua
N1 - Publisher Copyright:
© 2025 Elsevier Ltd
PY - 2026/1
Y1 - 2026/1
N2 - Coronary artery segmentation is a critical yet challenging task in diagnosing and evaluating coronary artery disease (CAD), as manual delineation is labor-intensive, time-consuming, and prone to inter-observer variability. Automated segmentation methods have been developed to mitigate these issues; however, existing approaches are often insufficiently accurate to meet clinical use standards. This is due to the intricate anatomical structures of coronary arteries (CAs), their sparsity, and the low contrast with adjacent tissues. To address these challenges, we propose a novel 3D deep network, FG-Net, designed to achieve high-quality coronary artery segmentation. Specifically, we introduce a novel boundary completeness enhancement (BCE) module for uncertainty reduction via fuzzy logic modeling, capable of enhancing the completeness of vessel boundaries and thus improving the network's ability to determine CA boundaries under challenging circumstances. Furthermore, we present a trajectory continuity enhancement (TCE) module based on a dynamic graph convolutional network (GCN) that is capable of adaptively enhancing the continuity of vessel trajectories, thereby improving the network's ability to deal with complicated artery structures spanning across diverse scales. Extensive experiments on two in-house datasets and three publicly available datasets consistently demonstrate that FG-Net achieves superior segmentation performance and generalization ability, outperforming multiple state-of-the-art algorithms across various metrics. Code is available at FG-Net.
AB - Coronary artery segmentation is a critical yet challenging task in diagnosing and evaluating coronary artery disease (CAD), as manual delineation is labor-intensive, time-consuming, and prone to inter-observer variability. Automated segmentation methods have been developed to mitigate these issues; however, existing approaches are often insufficiently accurate to meet clinical use standards. This is due to the intricate anatomical structures of coronary arteries (CAs), their sparsity, and the low contrast with adjacent tissues. To address these challenges, we propose a novel 3D deep network, FG-Net, designed to achieve high-quality coronary artery segmentation. Specifically, we introduce a novel boundary completeness enhancement (BCE) module for uncertainty reduction via fuzzy logic modeling, capable of enhancing the completeness of vessel boundaries and thus improving the network's ability to determine CA boundaries under challenging circumstances. Furthermore, we present a trajectory continuity enhancement (TCE) module based on a dynamic graph convolutional network (GCN) that is capable of adaptively enhancing the continuity of vessel trajectories, thereby improving the network's ability to deal with complicated artery structures spanning across diverse scales. Extensive experiments on two in-house datasets and three publicly available datasets consistently demonstrate that FG-Net achieves superior segmentation performance and generalization ability, outperforming multiple state-of-the-art algorithms across various metrics. Code is available at FG-Net.
KW - Coronary artery segmentation
KW - Feature enhancement
KW - Fuzzy logic
KW - Graph convolutional network
UR - https://www.scopus.com/pages/publications/105007060618
U2 - 10.1016/j.patcog.2025.111891
DO - 10.1016/j.patcog.2025.111891
M3 - 文章
AN - SCOPUS:105007060618
SN - 0031-3203
VL - 169
JO - Pattern Recognition
JF - Pattern Recognition
M1 - 111891
ER -