Topology-preserving coronary vessel segmentation in X-ray angiography using ResNet50-MASAG-UNet.
- 1School of Computer Science Engineering and Information Systems (SCORE), Vellore Institute of Technology, Vellore, India.
- 2Centre For Artificial Intelligence Research (C-FAIR), Vellore Institute of Technology, Vellore, India.
Abstract
Coronary artery disease occurs when the accumulation of fatty atheroma deposits narrows the coronary blood vessels, thereby reducing blood flow to the myocardium. To identify these narrowings, cardiologists use X-ray coronary angiography (XCA). However, these angiographic images are often blurry and noisy, making vessels difficult to view clearly. To enhance visibility, we proposed an image improvement technique that combines three filtering approaches: Bilateral filtering, unsharp masking, and Contrast Limited Adaptive Histogram Equalization (CLAHE), to remove noise, sharpen vessel edges, and improve brightness balance, producing clearer vessel images to support subsequent analysis. Our proposed ResNet50-MASAG-UNet integrates a Multi-Scale Adaptive Spatial Attention Gate (MASAG) at each skip connection of a ResNet50-UNet, enabling the decoder to attend selectively to coronary vessel regions while suppressing non-vascular activations from catheters and bones. A topology-aware loss is used to preserve vessel connectivity by maintaining centreline continuity. Together, these elements enable more accurate and complete coronary vessel segmentation. The proposed model achieves improved segmentation performance on XCA images, attaining a Dice coefficient of 88.3%, a Jaccard Index of 79.1%, a precision of 86.4%, and a recall of 90.5%. It outperforms the standard UNet baseline and provides clearer vessel segmentation, demonstrating improved potential for coronary vessel analysis.
Related
- Evaluating the real-world accessibility of open-source deep learning models for retinal vessel segmentation in optical coherence tomography angiography
- Nonlinear and Vessel-Specific Association Between Coronary Artery Calcium Burden and Coronary Computed Tomography Angiography Underestimation of Coronary Stenosis
- From coronary computed tomography angiography to the cath lab: computed tomography-guided percutaneous coronary intervention of a high-risk left anterior descending artery lesion
- Inflammatory biomarkers show sex-specific associations with cardiometabolic disease and coronary atherosclerosis as assessed by coronary CT angiography
- Assessing Calcified Coronary Lesions to Guide Percutaneous Coronary Intervention: From Angiography to Intracoronary Imaging
BibTeX
@article{sunilkumar2026topologypreserving,
title = {{Topology-preserving coronary vessel segmentation in X-ray angiography using ResNet50-MASAG-UNet.}},
author = {G Sunilkumar and P Kumaresan},
journal = {Frontiers in radiology},
year = {2026},
month = oct,
doi = {10.3389/fradi.2026.1949450},
url = {https://europepmc.org/article/MED/42953072},
note = {CC BY},
}
RIS
TY - JOUR
TI - Topology-preserving coronary vessel segmentation in X-ray angiography using ResNet50-MASAG-UNet.
AU - G Sunilkumar
AU - P Kumaresan
PY - 2026
DA - 2026/10/14
JO - Frontiers in radiology
DO - 10.3389/fradi.2026.1949450
UR - https://europepmc.org/article/MED/42953072
AB - Coronary artery disease occurs when the accumulation of fatty atheroma deposits narrows the coronary blood vessels, thereby reducing blood flow to the myocardium. To identify these narrowings, cardiologists use X-ray coronary angiography (XCA). However, these angiographic images are often blurry and noisy, making vessels difficult to view clearly. To enhance visibility, we proposed an image improvement technique that combines three filtering approaches: Bilateral filtering, unsharp masking, and Contrast Limited Adaptive Histogram Equalization (CLAHE), to remove noise, sharpen vessel edges, and improve brightness balance, producing clearer vessel images to support subsequent analysis. Our proposed ResNet50-MASAG-UNet integrates a Multi-Scale Adaptive Spatial Attention Gate (MASAG) at each skip connection of a ResNet50-UNet, enabling the decoder to attend selectively to coronary vessel regions while suppressing non-vascular activations from catheters and bones. A topology-aware loss is used to preserve vessel connectivity by maintaining centreline continuity. Together, these elements enable more accurate and complete coronary vessel segmentation. The proposed model achieves improved segmentation performance on XCA images, attaining a Dice coefficient of 88.3%, a Jaccard Index of 79.1%, a precision of 86.4%, and a recall of 90.5%. It outperforms the standard UNet baseline and provides clearer vessel segmentation, demonstrating improved potential for coronary vessel analysis.
SN - 2673-8740
ER -