BRANCH-BASED REGION GROWING METHOD
FOR BLOOD VESSEL SEGMENTATION
S.Eiho*, H.Sekiguchi *, N.Sugimoto?, T.Hanakawa ^, S.Urayama®
* Graduate School of Informatics, Kyoto University, Gokasho, Uji-shi, 611-0011 Japan —
eiho@image.kuass.kyoto-u.ac.jp
° Human Brain Research Center, Graduate School of Medicine, Kyoto University, Shogoin Sakyo-ku, Kyoto-shi,
606-8507 Japan
KEY WORDS: Segmentation, MRA, MIP, Blood Vessel, Region Growing
ABSTRACT:
We propose an algorithm of blood vessel segmentation for MRA data in this paper. The generic region growing, as well as
thresholding, is not appropriate to extract the whole part of the vessels on MRA data. This is because of the image property of the
MRA, where the intensity of each pixel on the blood area depends on the amount of blood flow. Moreover, thin vessels are affected
by the partial volume effect which reduces the intensity of vessel parts as the low pass filtering effect. So the range of the intensity
of the blood vessel in MRA image is not restricted in a small interval but spread widely. To get correct segmentation results by
region growing, the growing condition should be flexibly adapted according to the local characteristics in each ROI. We have
designed a branch-based region growing for this purpose. Since its growing process is performed on one branch at a time, the
growing conditions can be optimized according to its surrounding properties. It is also possible to connect a break point by
extending the vessel, which improves segmentation results. By applying this method to 5 head MRA data sets, the availability of the
method has been confirmed. In addition, to evaluate the segmentation result quantitatively, we developed a new evaluation method
which utilizes MIP data.
1. INTRODUCTION
|. Creating 3-D image requires blood vessel segmentation.
MRA (magnetic resonance angiography) and CTA (X-ray CT 2. Blood vessel segmentation on MRA data is quite difficult.
angiography) are widely used for the diagnosis of serious
circulation diseases. These data are basically slice images, and If blood vessel segmentation is easily obtainable, 3-D image
it is difficult to understand vessel's shape and their perspective will be available on MRA diagnosis. MRA is the only imaging
locations on the slice data. This is the reason why MIP method of blood circulation without invasiveness, and so, it is
(maximum intensity projection) or 3-D image is used for strongly desired to realize the blood vessel segmentation for
diagnosis. These images are created by accumulating a lot of MRA data.
MRA/CTA slice images.
Figure 1 shows an example of MIP and 3-D image. Both images 2. METHODS
are created from the same MRA data, but features and image
quality of them are quite different. In general, 3-D image is 2.1 Problems in the conventional methods
superior to MIP in regard to both reality and perspective.
As the range of the blood vessel intensity in MRA is widely
Ín recent years, 3-D images are commonly used for CTA spread, conventional binarizing method is unable to extract
diagnosis and a lot of 3-D applications have been developed. blood vessel region. The same is true for region growing,
On the contrary, MIP is mainly used for MRA diagnosis. because the growing condition is also determined from the
range of the intensity value. In addition, when the growing
proceeds in the narrow and long vessels, it often stops the
growing on the way because of noise or insufficient resolution
of the image.
To overcome this problem, it is obvious that the growing
condition has to be changed adaptively according to the local
vessel intensity. But a conventional region growing has several
growing points simultaneously as shown in Figure 2, and it can
hold one growing condition at a time. To solve the problem, we
proposed a new kind of region growing which keeps spreading
in restriction along only one vessel. Hereinafter we call it
"branch-based region growing".
Figure 1. Comparison of MIP (left) and 3-D image (right).
The reasons why 3-D images are not used for MRA data are:
796
