Graphical knowledge acquisition for medical diagnostic expert systems

Knowledge reuse among diagnostic problem-solving methods in the Shell-Kit D3

“…The results were quite encouraging: In a prospective evaluation with 51 unselected outpatients coming from a second clinic, it stated the final clinical diagnoses (which was used as gold standard) in about 90% of the cases and focused on the correct diagnoses in about 80% (i.e. in about 90% of the 90% cases with the final diagnoses) (Gappa et al, 1993). Since more than 1000 real cases were collected during development and evaluation time, case-based knowledge about a similarity measure are added to RHEUMA to exploit these data.…”

“…Thereby we emphasize the aspects of knowledge restructuring. D3 offers most of the above-mentioned methods including categorical, heuristic, statistic, set-covering, functional and case-based diagnostic problem-solving methods, and a complete graphical knowledge acquisition interface (Gappa, Puppe & Schewe, 1993). Knowledge in D3 can be used for various purposes.…”

“…Successful self-acquisition of knowledge by domain experts in the above mentioned projects was made possible by D3's graphical knowledge acquisition component using graphical aggregation and abstraction techniques for managing the complexity of large knowledge bases (Gappa et al, 1993;Bamberger et al, 1997). It not only simplifies the process of knowledge editing, but also makes the represented knowledge more transparent to the experts.…”

A comparison of two approaches to model-based knowledge acquisition

“…While new sub-methods must be explicitly coded, the corresponding knowledge acquisition components can be automatically generated from a declarative description of the knowledge representations and editors. This knowledgeacquisition tools allow domain experts after a training phase to develop the knowledge base by themselves without further guidance by a knowledge engineer [GPS93]. The experts may directly evaluate the knowledge base with test cases without the need for a further precompilation step.…”

“…The knowledge acquisition tools can be generated from a declarative description of the knowledge representation and the views acquired in the knowledge editors. The library of CRLM currently contain mechanisms for heuristic ([PuG92], [GPS93]) case-based [PuG91] and set-covering classification [Pup93], methods for assignment (propose-andexchange [PoP92]) and for simple configuration (propose-and-revise [PFL+94]). In MIKE, reuse is enabled by the very-high level language KARL which makes it possible to quickly write powerful (though inefficient) specifications of problemsolving methods and domain knowledge.…”