Information Technology Reference
In-Depth Information
2 The Classification Problem
One of the main difficulties in developing universal, intelligent systems for medi-
cal image diagnostics is the huge variety of forms of images, both healthy and
pathological, which have to be taken into account when supporting physicians that
interpret them. For this reason, the analysis should be made independent from the
orientation and location of the examined structure within the image. In addition,
every healthy person has an individual structure of their internal organs, everyone
is somewhat different, which prohibits us from unambiguously and strictly saying
what a given organ should look like, as it may look different and still be healthy
(i.e. fall within the so-called physiological forms). In particular, the aforemen-
tioned varied shapes of morphological elements make it difficult to set a universal
standard defining the model shape of a healthy organ, or a pathological one. All of
this means that attempts to effectively assess the morphology using computer
software are very complicated and frequently outright impossible, because there
are too many cases that would have to be analysed to unambiguously determine
the condition of the structure being examined. Neither do classical methods of im-
age recognition (i.e. simple classification) [9][18] produce satisfactory results - a
comprehensive analysis, the complete recognition and interpretation of the disease
symptoms looked for -- every time when supporting medical diagnostics (fig. 3.).
Fig. 3 A simple pattern classification methods rely strongly on quantitative measurements
and are not well qualified for all problems
Consequently, it becomes necessary to introduce a somewhat more advanced
reasoning aimed at recognising lesions and interpreting their meanings. The au-
thor's research shows that mathematical linguistic formalisms, and in particular
graph image grammars [12][23], can be quite successful in this area. However, us-
ing mathematical linguistic formalisms in the form of graph image grammars is
not free of shortcomings, either. One of the main difficulties is the need to define
the linguistic apparatus, i.e. develop a grammar so that there are deterministic syn-
tax analysers for it which will allow the lesions looked for to be recognised
[3][11][20][22]. Since, as a rule, it is very difficult to define some ideal, universal
pattern, e.g. an image showing some model shape of a healthy or diseased organ,
we are dealing with a situation in which we cannot define a complete and at the
same time finite set containing all possible forms of the pathology that can occur.
In addition, when there is a huge variety of shapes of the structures identified for
the purpose of their proper recognition (classification), it may become necessary
Search WWH ::




Custom Search