Information Technology Reference
In-Depth Information
to define a grammar very broad in terms of the number of productions introduced.
However, this problem can be solved by using grammars with greater generating
capacities. Still, one has to remember that for some grammars of this type there
may be problems with building deterministic syntax analysers. On the other hand,
a computer using the well-known and frequently used techniques of automatic im-
age recognition needs such a pattern to be provided to it. This is because the
information technologies applied rely to a significant extent on intuition to deter-
mine the measure of similarity between the currently considered case and such an
abstract pattern, so if the shapes of examined organs change unexpectedly as a re-
sult of a disease or individual differences, these technologies often fail. For this
reason it is necessary to use advanced artificial intelligence techniques and com-
putational intelligence techniques that can generalise the recorded image patterns.
What is particularly important is to use intelligent description methods for medical
images that would ignore the individual characteristics of the patient examined
and the characteristics dependent on the specific form of the disease unit consid-
ered. Linguistic descriptions of this type, created using new image grammars
modelling the shapes of healthy coronary vascularisation and the morphology of
lesions, form the subject of the rest of this publication.
3 Stages in the Analysis of CT Images under a Structural
Approach Utilising Graph Techniques
As 3D reconstructions of the coronary vascularisation can come in many shapes
and be presented in various projections (angles of observation), modelling such ar-
teries requires using suitably advanced description and classification techniques.
One of such techniques consists in image languages based on tree and graph for-
malisms [3][16][18]. Further down in this chapter, these very methods will be
used to discuss the basic stages in the analysis and recognition (classification) of
lesions in CT scans of coronary vascularisation.
Under the syntactic (structural) approach [3][11][16][18][20][21][22], a com-
plex image is treated as a hierarchical structure made up of simpler sub-images
which can be broken down into even simpler ones until we get down to picture
primitives. Then, depending on the relations between these primitives and using
the appropriate formal grammar, the structure of the image can be represented as a
series, a tree or a graph. Defining and identifying simple components of an image
makes it possible to describe the shape of the analysed structure, and then to create
a generalised, holistic description defining the shapes of lesions looked for in the
item analysed. Thus, for the given image (showing a healthy structure or lesions),
we obtain certain significant sequences describing the lesions of interest to us.
Then, the process of recognising (classifying) the image boils down to a syntac-
tic/semantic analysis (parsing) whose purpose is to identify whether the analysed
input series is an element of the language generated by the grammar (fig. 4.).
Search WWH ::
Custom Search