Information Technology Reference
In-Depth Information
operate using their particular guidelines and procedures, (2) many clinical rules
must be modified to be applicable to specific groups of patients, and (3) some clini-
cal rules must be simplified to eliminate unavailable test results. To provide for the
adaptability of KB, we have added a few adaptation rules. However, this aspect of
the KB maintenance relies mostly on the user-driven updates to the rules, and has
not been fully automated.
Learnability of KB:
The users of CDSS expect the KB to be able to learn from the
experience. To provide for the learnability of KB, we have used two components:
data warehouse of medical cases and data mining tools. Since both of these com-
ponents are also used to support the V&V process, we describe them in the next
section.
14.4
Verification and Validation of the KB
In this section, we discuss verification and validation of the knowledge bases for
CDDS. First, we differentiate between verification and validation. Second, we de-
scribe verification and validation process using examples from two CDDS, which
support diagnostic tasks in two medical fields: psychiatry and sleep medicine. Both
CDDS systems are based on fuzzy rules, and both support relatively narrow diag-
nostic tasks. The first system, which supports diagnosis of clinical depression, uses
expert knowledge and clinical guidelines for the creation of the KB. The second
system, which supports the diagnosis of obstructive sleep apnea, uses a combina-
tion of the medical literature and clinical experts' knowledge for the creation of the
KB. We use these two systems to demonstrate the built-in support for verification
and validation.
14.4.1
Verification vs. Validation
In software engineering, verification means checking if the software system is built
according to the specification. On the other hand, validation means that the built
software system is “valid” in the context of the user problem. Verification is a pro-
cess which controls the quality and is used to determine whether the software system
meets the expected standards. In contrast to this, validation is a process which as-
sures the quality. It gives an assurance that the software system is successful in
accomplishing what it is intended to do (solves the user problem).
The verification and validation (V&V) of the knowledge base is different from
the V&V process for a typical information system. For example, in a rule-based
system, verification is the process of checking the “syntactical” correctness of the
rules, whereas validation is the process of checking the “semantic” and “pragmatic”
correctness of the rules. A rule is semantically correct, if it is clinically valid. For
example, a clinical prediction rule or diagnostic criteria must be validated through
a number of research studies and approved by medical associations. A rule is prag-
matically valid, if it is useful in a particular clinical setting. For example, a rule
