Information Technology Reference
In-Depth Information
2.4.2 Semantic Analysis in Medical Systems for Cognitive Data
Interpretation
Semantic categorisation need not always apply to speech, as there are classes of
topics which, when subjected to semantic analyses, splendidly illustrate its cogni-
tive potential. The authors have spent several years researching the use of seman-
tic analysis for the cognitive categorisation of,
inter alia
, images of the central
nervous system including changes in the appearance of the spinal cord, lesions of
foot bones and of palm bones. The purpose of the medical image analysis con-
ducted was to find what deformations have occurred in the examined organ and
what disease processes they may be symptoms of. Just recognising the analysed
shape and an attempt to classify it are not enough, as these can very easily be dis-
turbed, even if only by the natural differences between individuals concerning the
structure of the organs examined, and also if there is no lesion present. With
regards to cases of an organ in which there actually is a lesion, the analysis con-
ducted is aimed at confirming the occurring pathology of this organ and identify-
ing other information significant for both the physician and the patient.
The semantic analysis of medical images, just as speech analysis, suffers from
some difficulties, one of which is certainly the lack of a universal pattern of an
ideally healthy human organ due to the varied anatomical structure of the organ in
various people. Creating a database containing a certain number of perfectly de-
fined patterns does not mean that all possible situations (pathological cases) have
been classified and their representations included in the system base. Conse-
quently, one has to account for the fact that the same disorders cause deformations
differing in size and shape and in addition occurring in different places within the
examined organ, and for situations in which (apparently) very similar changes in
the shape of the organ image are associated with completely different clinical in-
terpretations.
Due to the possible difficulties that can occur in the diagnostic interpretation of
each analysed medical image, it can justly be said that the recognition itself of le-
sions in the analysed image is not sufficient to attempt a diagnosis, whereas at-
tempting to understand the image in the sense of determining the nature of the
process that led to this and not another deformation of the organs visible in the
image offers much better opportunities for an accurate interpretation. This is why
it makes sense to make the automatic data analysis process go beyond a superficial
analysis of the image form and an attempt to classify it, and instead attempt to un-
derstand the image. Machine understanding of the analysed medical image can be
used as an example of a task requiring in-depth, cognitive data analysis. The entire
process leading to understanding the image is extremely complex and its individ-
ual phases are inseparable from and interdependent on one another. This process is
illustrated in Figure 2.21, beginning with the pre-processing stage, showing its in-
dividual elements and the geometric transformations aimed at simplifying the rep-
resentation (e.g. the straightening transform).
