(Projektbeschreibung nur auf Englisch)
K. Boegl, G. Grabner, K.-P. Adlassnig
Department of Medical Computer Sciences, Section on Medical Expert and Knowledge-Based
Systems, University of Vienna Medical School, Spitalgasse 23, A-1090 Vienna, Austria
e-mail: karl.boegl@univie.ac.at
Background
The acquisition of medical knowledge and the knowledge representation framework of a
medical expert system are closely related to each other. We have experienced in the
past, that the crucial point of knowledge base development is to provide an intuitive formalism
of how to transfer human expertise into a systematized and structured form as
required by computer expert systems.
Objective
The aim of this project was to explore new ways of knowledge engineering that are exemplified
on liver diseases (hepatology) [1]. Special emphasis was laid on the determination of the
strength of confirmation of a present symptom for establishing a disease as
diagnostic hypothesis. Variants of the positive predictive value are calculated for this purpose.
Other topics of interest were disease hierarchies, cooperating knowledge bases,
and systemic interrelationships among symptoms, signs, test results, and diagnoses.
Material and Methods
Based on statistical (Bayesian), frequentistic, and fuzzy logical models of knowledge
representation and reasoning this project aimed at the development of a methodological
and mathematical framework that could be used for future medical expert system development.
The project was performed in close cooperation between an experienced clinician and
computer scientists. Some major problems that were dealt with specifically were
Several mathematical models were developed to facilitate the knowledge acquisition process. As an example, a formalism was developed that allows an interpretation of the strength of confirmation value as a frequentistic interpretation of probability with and without considering the prevalence of the diseases [2,3]. Based on this formalism a semiautomatic computation of the strength of confirmation values from the inverse frequency of occurrence values is feasible.
Results
At present the knowledge base covers 125 classical hepatological diseases and about
950 symptoms, signs, and test results. A pool of typical hepatological disease patterns
(35 hypothetical cases) was established to allow a pre-clinical evaluation of the knowledge bases.
Technical Specification
Most of the pre-clinical testing was performed by Microsoft Excel 97 Macros. An early
prototype, named HEPexpert, was realized in Java (Java Development Kit (JDK) 1.1.5) to
allow a comprehensive evaluation in a clinical environment (Fig. 3).
Conclusion
Our results showed that the development of large knowledge bases, as required for comprehensive
clinical domains such as hepatology, can be essentially facilitated by employing intuitive
knowledge representation formalisms. The use of fuzzy set theory extends the scope of the
formalisms to vague and uncertain knowledge.
References