h1

%0 Thesis
%A Liem, Steve Wei-Lung
%T Semantische Modellierung für ein wissensbasiertes System in der Pädiatrie
%C Aachen
%I Publikationsserver der RWTH Aachen University
%M RWTH-CONV-112871
%P II, 97 S. : Ill., graph. Darst.
%D 2008
%Z Aachen, Techn. Hochsch., Diss., 2008
%X This project aimed to create a basic framework for a computer-based system, which enables physicians and medical students to quickly and efficiently retrieve medical knowledge. The knowledge base DataMed is designed to assist the physician with finding diagnoses and making clinical decisions in his or her daily routine. Query algorithms make complex queries possible, which conventional medical literature does not allow. Textbooks have a linear structure and are ordered by chapters (diseases). They are principally suitable to comprehensively present knowledge about a disease. DataMed can also be used as a reference. Its advantage over conventional textbooks lies in its capacity to reach a possible diagnosis with the use of incomplete information, e.g. a small set of symptoms. This is the result of the semantic network, with which information about different diseases is interlinked and is now accessible to computer-aided processing. The main focus of this work lies in the conception of the data model. The first step was the selection of a suitable knowledge domain. The domain of pediatric viral diseases was chosen, because it is representative and manageable. Its diseases are characterised by categories like etiology, symptoms or therapy, and are therefore representative for many internal diseases. In addition, there are intersections between the symptoms of the diseases, which allows a comparative approach with differential diagnoses. The modeling of knowledge was carried out under the paradigm of object orientation, weak typisation, and the semantic network. The textbook information was analysed, and translated into a semantic network. This consists of objects and links. An assertion like “measles cause fever” is broken down into the objects “measles” and “fever”. These two objects are then connected by a link with the name “has symptom”. In order to produce a valid semantic network, the model was based on a well-established nomenclature, the SNOMED. The hierarchy of the SNOMED ensures consistency and validity in case of future expansions. The current DataMed model consists of 549 concepts, 1041 associations, 183 container objects and 214 SNOMED-concepts. The conception of the data model was initially done independent from the computer. Modeling steps alternated with evaluation steps even before implementation. Improvements and “best practices” in challenging cases (e.g. symptoms with characteristic) were successfully integrated into the data model. It is object oriented and has a high granularity. Objects in the model are kept as simple as possible and follow the principle of weak typisation. Most information about the objects is inherent in the semantic network, consisting of objects and associations. After the model was consistent enough, it was digitised with a UML-editor (Poseidon). With an import module, the UML model was mapped into a object oriented databas (FastObjects). The quality of the data model was then evaluated in example sessions and a comparative analysis of original text and data model. Currently, the model only contains a small sample of medical knowledge. However, the model represents a stable and efficient basis for larger medical knowledge domains. It is consistent and capable of incorporating future expansions in form and content. The modeling approach is, as textual analysis has shown, not capable of entirely mapping medical knowledge. Medical texts rely heavily on impicit information hidden in sentence structure of natural language. The ratio of modeled information can be increased through extensions. The abstract nature of the object oriented model can be complemented through integration of full texts and additional multi-media elements (images, audio, and video). Computer-based knowledge processing is quickly gaining importance in the field of medicine. It was the objective of this work to present criteria and solutions, which can be used to translate a conventional medical text into an abstract data model, which is then integrated into a concrete medical knowledge base for clinical use. It shows the feasibility of correctly and didactically representing medical knowledge for a computer-based system. With DataMed, problems and their solutions were identified, which resulted from the dilemma of translation between natural language and machine-readable information.
%K Semantisches Netz (SWD)
%K Objektorientierung (SWD)
%K Semantische Modellierung (SWD)
%K Wissensrepräsentation (SWD)
%K SNOMED (SWD)
%K Kinderheilkunde (SWD)
%K Wissensbasiertes System (SWD)
%F PUB:(DE-HGF)11
%9 Dissertation / PhD Thesis
%U https://publications.rwth-aachen.de/record/50322