A NEW METHOD OF VESSEL CENTERLINE EXTRACTION
FROM 3D CT CORONARY ANGIOGRAPHY BASED ON
OPEN-SNAKE
Di Zhang *, Cheng Wang *, Shoujun Zhou
†
* Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences ,China ,email:1023516026@qq.com
†
Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences ,China ,email:sj.zhou@siat.ac.cn
Keywords: Vessel centerline, GVF, open-snake, CTCA.
Abstract
Efficiently obtaining a reliable coronary artery centerline from
computed tomography coronary angiography (CTCA) data is
relevant in clinical practice. In this paper, open-snake is
presented to extract the vessel centerline, which is drove by
two external forces, one is Gradient Vector Flow (GVF), and
the other one is an adaptive stretching force acted on the two
ends of open-snake. To make the open-snake working
effectively in the CTCA data, the following steps are used to
pre-process, initialize and control the deforming process.
Firstly, the multi-scale filter is used to enhance vessels of the
CTCA data, based on which ridge-point tracking is
implemented to initialize the open-snake. When the stretching
force is acted on the two ends, the open-snake is adaptively
deforming along the vessel’s ridge-lines. The process is
stopped until the deforming curves are not changed any longer.
As a semi-automated method, the open-snake can evolve along
the centerline of the vessel with little user interaction. In the
experiments, standardized evaluation methodology is used to
measure the extraction effects of coronary artery centerline.
The results indicate that our method can extract most of the
coronary artery centerlines with a good performance.
1 Introduction
Computed tomography coronary angiography is widely used in
clinical routine for coronary artery studies. Extracting
centerlines of the coronary artery is important in the coronary
related clinical applications. However, vessel segmentation is
still an open problem with respect to some factors such as
image modality, the required interaction mode and so on.
A lot of methods have been proposed in the literatures [1, 3].
Among these methods, parametric active contour mode (snakes)
was firstly described by Kass et al. [2], which provided a
solution to detect the contours of objects. Then, Snake was
introduced into medical image processing quickly. McInerney
et al. [6] and Mille et al. [7] used Snake to extract contours of
vessels in 2D image, however the method could become very
complex in 3D image. In recent years, Li et al. [4, 5] and Xu et
al. [10] used open active contours (open-snake) to extract the
centerlines of actin filaments in 3D image, and acquired good
effect on actin filaments segmentation.
For coronary artery extraction, the above-mentioned open-
snake method will be largely limited with complex anatomy
surroundings such as pulmonary artery,aorta and ventricle
edges. Thus, a new open-snake based method is proposed in
this paper to extract coronary artery centerline. Our method has
some optimal strategies: On one hand, we improve the open-
snake with a novel external force; on the other hand, we
propose an affective ridge-points tracking (RPT) algorithm to
initialize the open-snake. The proposed method can extract
most centerlines of the main branches of the coronary artery
automatically with little user interaction (just the start point
around the coronary artery ostia).
2 Methods
Our method includes three steps: firstly, the CTCA date is
processed using multiscale filtering to improve vessel structure
and suppress the non-vessel noise. Then, the initial curves of
vessel centerline is acquired based on RPT algorithm. Next, the
open-snake curves with adaptive stretching forces on its two
ends is produced based on the initial curves, which can deform
and approximate the real vessel centerline according to
principle of the energy minimization.
2.1 Open-snake
Open-snake is open-ended parametric active contours. When
two stretching forces are applied at both ends of the open curve,
the models can elongate while conforming to the desired image
feature. The illustration of open-snake is shown in Figure 1.
Figure 1: Illustration of open-snake. Two stretching forces are
applied at two ends, respectively. The image’s gradient vectors
of some curve points are shown using the red arrows, which
pushes curve points move toward the vessel centerline.