Information Technology Reference
In-Depth Information
requiring complex laboratory tests may not be suitable for use in a clinic not
equipped with a highly specialized laboratory equipment. In many cases, the veri-
fication process can be automated; the validation process must be performed by the
users. However, the users can be assisted by a computerized analysis of data.
14.4.2
Rule-Based Representation for KB
Our examples are based on two CDSS systems, in which rules have three functions:
descriptive, prescriptive, and predictive. In the descriptive sense, rules characterize
the subpopulations of patients with higher or lower risks for the disease. In the pre-
scriptive sense, rules define the typical (normative) values for specific predictors. In
the predictive sense, rules assess the probability of a new patient belonging to one
of the classes. The hypothetical quality of the rule is defined by the certainty factor
(CF), a degree of belief ranging from
0 (absolute
belief), assigned to the rule by medical experts based on their clinical experience.
Similarity to MYCIN, we use certainty factors (CF) to represent uncertainty char-
acteristic to medical application. Furthermore, we use a fuzzy-logic approach to
represent impression. Thus, the rules can be crisp and fuzzy.
The rule is comprised of two parts: a premise and a consequent. The premise of
the rule uses predefined predictors, for example, age, gender, or snoring. A proposi-
tion is a logical expression composed of a predictor variable, the relational operator
(
−
1
.
0 (absolute disbelief) to
+
1
.
<
,
≤
,
>
,
≥
,
=
), and a value; for example, age
>
65, habitual snoring
=
yes. The
>
rules are in the conjunctive propositional form, for example, age
65 AND gender
=
female. The conclusion of the rule includes the class label. The following two
rules are simplified examples of diagnostic rules used in the screening for obstruc-
tive sleep apnea (OSA):
R1:
IF habitual snoring
=
yes THEN OSA
=
yes (CF
=
0
.
4)
R2:
IF habitual snoring
=
yes AND age
=
older THEN OSA
=
yes (CF
=
0
.
5)
The first rule, R1, has one binary predictor: habitual snoring (yes/no). Habitual
snoring is one of the most important predictors of OSA. However, there are cases
of OSA without snoring, as well as there are patient who habitually snore and do
not have OSA. Studies have shown [4] that approximately 50% of habitual snorers
have some degree of sleep-disordered breathing. Therefore, R1 has certainty factor
of 0
4. The second rule, R2, has an additional predictor: age. Age is a linguistic
variable. Older age increases the chances of having OSA. On the other hand, older
age increases the chances of snoring. Therefore, R2 has certainty factor of 0.5.
.
14.4.3
Verification of KB
Knowledge bases for medical applications, even for narrowly specialized DSSs,
are large and complex. Therefore, on-going verification of such systems requires
