Expertensystem Anämiediagnostik

Summary The aim of this thesis was to develop a knowledge-based expert system for the diagnosis of anaemia. The software j.MD - Java system to Support Medical Diagnostic Reasoning - was provided by the company Wormek in cooperation with Sysmex Germany. Building on universally valid explanations for the diagnosis of anaemia (such as morphological criteria and aetiologic-pathogenetic viewpoints) that formed the basis for the set of rules in the j.MD knowledge base, decision trees where constructed. The production of an expert system, with the j.MD software on the predecessors model Pro.M.D., is the central segment of the present work. Three subranges of the j.MD software – production and testing of knowledge bases (j.MD-Knowledge Editor with j.MD Debugger) as well as the separate user interface (j.MD Front End) – were introduced and examined. The j.MD Debugger is the validation tool of the knowledge-base and text-base, and was used to identify a faulty element which occurred when a mistake was present in a decision tree or the input patient data. While the Front End tool not only enables the user to quickly check the data but also generates the report immediately. This report arises from processing the links between the elements of the knowledge base and text base, according to the concepts and rules of the expert system. The report was presented in a defined, individually fixable structure. Of minor importance was the reports programming layout and structure. A knowledge base was provided to the anaemia diagnostics (red blood count) which consists of two main parts, the knowledge base and the text base of the expert system. All together both parts contain 76 concepts and subconcepts with 76 matching rules as well as 30 laboratory parameters. The suitable LOINC codes were assigned to the laboratory parameters. Additional rules were necessary to define layout and structure of the report in the text base. Data (concepts and rules) could only be tested if the debugger tool had patients data for its validation. Therefore 100 test patient sets with laboratory parameters were designed within j.MD, however, the test data had to be stored externally on the hard drive of the computer and are not visible in the j.MD programme. The results of the debugger data validation are summarised and shown in the j.MD Report. j.MD is a suitable software for the development of knowledge-based expert systems. However, for laymen the programming is very difficult, especially without introduction and accompanying experts support. In spite of an introduction seminar, the support of a so-called knowledge engineer was much appreciated as the software was not self-explanatory. As j.MD was written in JAVATM, the platform-independent computer language, it should have provided an easy integration into given laboratory computer environment. During construction and programming of the knowledge base, it was extremely difficult to deal with updates and new features for the software. At that time j.MD was not at end product stage and therefore bugs had to be eliminated. Errors also occurred with the integration of new updates which again had to be corrected by the company Wormek. The preloaded LOINC© data bank provided laboratory parameters, which allowed an exact coding of parameters and offered at the same time a connection with internal communication systems. The data bank is regularly updated by the Regenstief Institute and proved to be a very useful element as the laboratory parameters did not have to be defined anew. The aim to generate a knowledge-based expert system for the diagnosis of anaemia with the help of the j.MD software was achieved. On account of the various difficulties in usability and liability, as well as unexpected restrictions (for example, the fact that only unequivocal authoritative areas and only the red blood count could be introduced in the expert system), the integration of the j.MD expert system into the laboratory computer and analytical system in Marburg was deemed inappropriate.