Information Technology Reference
In-Depth Information
-
Influence diagrams make it possible to take decisions in a normative way to
establish the most appropriate action policy: complementary explorations,
pharmacological and cognitive treatments, even for cases where it is not
so obvious and the doctor's clinical judgement is unable to find the best
solution,
-
Decision analysis can explicitly and systematically combine different experts'
opinions and experimental data, such as data from studies published in
medical literature. [4]
-
There are a number of analyses applicable to Bayesian Networks and Influ-
ence Diagrams, which provide added value to the technique, like for example,
Evidence Conflict Analysis, Sensitivity Analysis (Evidence and Parameters),
Value Analysis of the source of information.
There are Bayesian Network Frameworks that perform all or some of these anal-
yses and they can be extremely useful for the problem that we are addressing,
because they can also continually validate the model and facilitate the discovery
of new relations, analyse new risk factors, new treatments, etc.
3 Proposal for the Diagnosis
As indicated earlier, the Bayesian Network consists of quantitative and qualita-
tive models. For problem modelling four types of variables are used:
-
Context information variables or risk factors. It is the information that is
present before the problem occurs and that has a causal effect on the prob-
lem. In this group of variables we have: age, sex and educational level.
- Information variables representing the symptoms. Variables representing
whether the patient has a conceptual-semantic-lexical deficit. These vari-
ables analyse the features or attributes contained in a number of semantic
category definitions of living or non-living creatures, and other specific ones
for each of these categories. The common ones are: taxonomical, functional,
part-all, evaluative, place/habitat, types, and the uncommon ones: proce-
dure, behavioural activity, cause/generation and origin. Attribute taxonomy,
which acts as a schema or theoretical and methodological evaluation frame-
work for this test, the same as in the second test, can be seen in detail in
Peraita, Elosúa and Linares (1992). These variables represent attribute pro-
duction in each semantic category of each of the objects used (apple, dog,
pine, car, chair and trousers).
-
Intermediate variables. They are variables that cannot be directly observed
whose a posteriori probabilities are not of immediate interest, but they play
an important role in achieving correct dependence and conditional inde-
pendence of the properties and therefore ecient inference. Intermediate
variables represent the semantic categories of living creatures and non-living
creatures, the semantic content deficit of the different categories, etc. In this
type of variables we have: Cognitive Deterioration Living and Non-Living
Creatures on the one hand, and Cognitive Deterioration Apple, Dog, Pine,
Chair and Trousers, on the other.
Search WWH ::
Custom Search