Project description
Coronary artery disease (CAD) is one the most common heart diseases and a leading cause of death worldwide, and X-ray coronary angiography is considered the gold standard for clinical decision-making and therapy guidance. Its interpretation, however, is manual, operator-dependent, and time-consuming. Automated analysis starts with vessel segmentation, which is challenging on X-ray angiograms due to low contrast, background artefacts, and overlapping, foreshortened vessels. Classical methods rely on hand-crafted vessel enhancement, matrix-decomposition/Hessian models and multiscale filtering with neural-network classification, while deep learning, especially U-Net variants, now enables robust, near-real-time vessel segmentation.
This project develops an end-to-end deep learning pipeline for complete coronary tree analysis, covering vessel segmentation, topological analysis, centerline extraction, guidewire detection, and stenosis localization/grading. The student can apply (and compare) a range of methods, from CNNs (U-Net/nnU-Net) and Transformers (TransUNet, SegFormer) to promptable foundation models (SAM/MedSAM), weighing their accuracy, annotation cost, and generalization.
Requirements
- Background in machine learning, deep learning, or computer vision.
- Working knowledge of Python and a deep learning framework (PyTorch preferred).
- Willingness to work with medical image data and topological/geometric analysis of vessel structures.
Keywords
Coronary angiography, X-ray, vessel segmentation, centerline extraction, topological analysis, guidewire detection, stenosis localization, CNN, Vision Transformer, Segment Anything Model (SAM), medical image analysis.
Supervisor team: Dr. Kenan Niu will be the supervisor. Dr. Sahar Nasirihaghighi will be the daily supervisor.
References
[1] Xia, Shaoyan and Zhu, Haogang and Liu, Xiaoli and Gong, Ming and Huang, Xiaoyong and Xu, Lei and Zhang, Hongjia and Guo, Jialong. Vessel segmentation of X-ray coronary angiographic image sequence. IEEE transactions on biomedical engineering, 2019. P 1338—1348.[2] Yang, Su and Kweon, Jihoon and Roh, Jae-Hyung and Lee, Jae-Hwan and Kang and others. Deep learning segmentation of major vessels in X-ray coronary angiography. Scientific reports, 2019.[3] Gao, Zijun and Wang, Lu and Soroushmehr, Reza and Wood, Alexander and Gryak, Jonathan and Nallamothu, Brahmajee and Najarian, Kayvan. Vessel segmentation for X-ray coronary angiography using ensemble methods with deep learning and filter-based features. BMC Medical Imaging, 2022.