Information Technology Reference
In-Depth Information
a)
b)
c)
Fig. 1. Sample heart image: (a) - sample slice, (b) - masks of right ventricle obtained
from contours drawn by an expert ( M ri,in -white, M ri,ou - black), (c) - masks of left
ventricle obtained from contours drawn by an expert ( M le,in -white, M le,ou -black)
decades the development of computed tomography (CT) allowed to introduce
new diagnostic methods of pulmonary embolism. Some of them were mentioned
in [4]. This paper focuses on a method for the right ventricle systolic pressure
assessment linked with shifts in interventricular septum curvature that requires
drawing of endocardial and epicardial contours. Currently they are drawn manu-
ally, which is time-consuming as well as prone to errors due to technical mistakes
and tiredness (Fig. 1). The aim of this work is to automate this process.
The analysed image data are obtained using ECG-gated computed tomogra-
phy scanner in a standard chest protocol after intravenous injection of contrast
media. Next, heart cycle is reconstructed in 10 phases and two chamber short
axis view is generated leading to 4D image sequence, i.e. the set of 2D images
for each reconstructed slice and for each phase of heart cycle.
3
Proposed Approach
The proposed method tries automatically to imitate the top-down process of
right and left heart ventricle contour localization. This process is divided into
three phases, i.e. preprocessing, segmentation and improvement phase that are
described below. It uses adaptive potential active contours (APAC) presented in
[4,5] with different formulation of energy function in each phase. The choice of
potential active contour approach resulted from specific properties of potential
contour model that, in a natural way, allows to describe smooth, medical shapes.
Other traditional active contour techniques such as snakes or geometric active
contours require special steps, e.g. definition of additional smoothing internal
energy components, in order to achieve similiar results. It should be emphasized
that the presented approach is currently not fully automatic because the expert
must choose, for each 4D image sequence, a proper threshold that allows to
distinguish in the images those areas that represent blood with injected contrast
from the rest of the image (Fig. 2a). This requirement, however, in comparison
with manual contour drawing is not a dicult task. Nevertheless, there are
also conducted works that aim at solving that problem. Below the following
denotations are used:
- I
- set of all pixels in the image;
Search WWH ::




Custom Search