14-08-2012, 03:42 PM
Segmentation of ultrasound images of the carotid using RANSAC and cubic splines
Segmentation of ultrasound images.pdf (Size: 668.65 KB / Downloads: 61)
Introduction
Non-invasive ultrasound imaging of human arteries is a
widely used form of medical diagnosis of arterial diseases,
like atherosclerosis, a disease of blood vessels caused by the
formation of plaques inside the arteries. Atherosclerosis is
quantitatively evaluated by the intima-media thickness (IMT),
which measures the distance between the inner boundary of
the adventitia and the lumen, the region of the vessel where
the blood flows. The IMT of extracranial carotid arteries, which
can be measured using B-mode imaging, provides an index of
individual atherosclerosis and is used for cardiovascular risk
assessment in clinical practice [1]. The diagnosis of atherosclerosis
is one of the most important medical examinations
for the prevention of cardiovascular events, like myocardial
infarction and stroke [2,3]. However, it requires the detection
of both the adventitia and the lumen boundaries.
Materials and methods
Dataset
A set of 50 longitudinal B-mode images of the CCA was
acquired with a Philips HDI 5000 ultrasound system and
recorded with 256 gray levels. Seven of these images include
a part of the internal carotid artery. The image pixel size was
normalized to 0.09mm, a common value used in clinical practice.
The parameter settings of the scanner were not kept the
same for every image since we aimed at achieving robustness
to different settings.
Edge estimation
The main goal of the edge estimation step is to obtain a map
of pixels whose properties, such as edge magnitude, gradient
orientation and valley shaped intensity profile, are compatible
with the adventitia boundaries. The estimation of edges can be
divided into the following main steps: edge detection (Section
2.3.1), estimation of the dominant gradient direction at edges
(Section 2.3.2), selection of the edges of interest that define the
final edge map (Section 2.3.3) and determination of the valley
edge map (Section 2.3.4).