11 Apr 2016 |
Research article |
Software Systems, Multimedia and Cybersecurity , Health Technologies
Vessel Walker: A new Approach for Coronary Arteries Segmentation





The segmentation of vascular structures is performed during vessel measurement, diagnosis and treatment planning. It is a highly challenging task due to the wide range of vessel sizes, shapes and intensities which is further complicated by the intricate topology including vessel bifurcations and overlap, and local deformations such as aneurysms or stenoses. The authors proposed a new solution, the vessel walker, which extends the Random Walker formulation by integrating vesselness information.
Introduction
According to the World Health Organization, cardiovascular diseases such as coronary heart disease are the first worldwide cause of death [1]. To diagnose and treat these diseases, minimally invasive percutaneous interventions like balloon angioplasty are often preferred over open-heart surgery, due to their shorter operation and recovery times, and their reduced risk of postoperative
complications. In balloon angioplasty, coronary vessels are revascularized by inserting a catheter in the affected coronary and inflating a balloon to put a stent in place. This intervention is
routinely conducted under biplane or monoplane fluoroscopic guidance, where a contrast agent such as iodine is injected at key moments to enhance the coronary vessel lumen diameter. The images produced by this process are known as X-ray angiographies.
One of the major challenges during navigation in the coronary arteries is that the contrast agent cannot be continuously injected in the targeted vessel because of its toxicity. To provide guidance under low contrast, coronary vessels can be outlined by segmentation. The segmentation of vascular structures, which can also be used for vessel measurement, diagnosis and treatment planning, is a highly challenging task due to the wide range of vessel sizes, shapes and intensities, to the complex topology including vessel bifurcations and overlap, and to local deformations such as aneurysms or stenoses.
Vessel-Like Structure Segmentation
The problem of segmenting vessel-like structures is well documented in the literature (e.g., see [2] for a comprehensive survey). Among the proposed solutions for this problem are vesselness filters [3, 4], which estimate the vessel centerness probability of pixels using the spectral properties of the Hessian matrix, at various scales. However, these filters may lead to disconnected regions and meet difficulties at local deformations to the vessels such as aneurysms or bifurcations. Graph-based methods, using minimal paths [5], random walks [6] and graph cuts [7, 8, 9], have also been proposed in the literature. In many of these approaches, like [7] and [8], intensity values are used as class priors of pixels or voxels. In angiograms, however, this information is unreliable since smaller vessels can have the same intensities as the background.
In this paper, we present a new vessel segmentation approach that extends the formulation of [6] by integrating vesselness information (see fig. 1 below). While recent works have also proposed to combine vesselness with graph cuts [8, 9], our approach offers several advantages. Thus, unlike [9] vesselness values that are thresholded to generate seeds, our method uses these values directly in the energy function, which makes it more efficient in differentiating the tubular structurrandom walkses in the background (false positives).
Moreover, our method may also use manually entered seeds to refine the segmentation of smaller vessels or noisy regions. In [8], seeds are only considered in a post-processing step and cannot be used to remove background noise. Finally, as mentioned in [6], methods based on random walks may be preferable in some cases, since they are less sensitive to the shrinking bias problem.

Fig. 2 Segmentation results obtained with automatic thresholding. From left to right: Original images, Active Contour results, Random Walker results and our proposed method using the fully automated and the semi-automatic formulations (Source [Img2])

Fig. 3 Foreground class probabilities obtained by the Frangi filter (left) and our proposed method (right (Source [Img3]).
Conclusion
We proposed a new interactive vessel segmentation method that extends the Random Walker formulation by integrating vesselness information. This method was tested on 2D X-ray coronary artery angiographies and obtained more accurate results than the active contour-based method. In future works, we will evaluate the effect of seed point selection on the segmentation performance, and investigate the usefulness of other types of vesselness priors.
This research was funded by a grant from the Fonds de recherche du Québec Nature et technologies FQRNT (http://www.frqnt.gouv.qc.ca/en/le-frqnt).
For more information
For more information, please see the following research paper: Vessel Walker: Coronary Arteries Segmentation using Random Walks and Hessian-based Vesselness Filter, source.

Faten M'Hiri
Faten M’hiri is a PhD student in the Software and IT Engineering department at ÉTS. She is a member of the Interventional Imaging Lab of ÉTS. She received a bachelor in computer science in Tunisia and a Masters in engineering at ETS.
Program : Software Engineering
Research laboratories : LIVE – Interventional Imaging Laboratory

Luc Duong
Luc Duong is a professor in the Software Engineering and IT Department at ÉTS, and researcher at the CHU Research Center. His research focuses on medical imaging, computer vision, algorithms and artificial intelligence.
Program : Software Engineering Information Technology Engineering
Research laboratories : LIVE – Interventional Imaging Laboratory

Christian Desrosiers
Christian Desrosiers is a professor in the Software Engineering and IT Department at ÉTS. His research interests include data mining, machine learning, biomedical imaging, recommendation systems and business intelligence.
Program : Software Engineering Information Technology Engineering
Research laboratories : LIVE – Interventional Imaging Laboratory LIVIA – Imaging, Vision and Artificial Intelligence Laboratory

Mohamed Cheriet
Mohamed Cheriet is a professor in the Department of Systems Engineering at ÉTS and Director of Synchromedia. His research focuses on eco-cloud computing, knowledge acquisition and artificial intelligence systems and learning algorithms.
Program : Automated Manufacturing Engineering
Research chair : Canada Research Chair in Smart Sustainable Eco-Cloud
Research laboratories : SYNCHROMEDIA – Multimedia Communication in Telepresence CIRODD- Centre interdisciplinaire de recherche en opérationnalisation du développement durable
Field(s) of expertise :
