Research Students Supervised: Past and Present

Elisabeth Bacon

The Design of Maintainable Knowledge Based Systems: A Case Study in the Domain of Magnetic Resonance Imaging

The problem of maintenance of expert systems is investigated using an example in the domain of magnetic resonance (MR) imaging. The term maintenance is used here in the sense of making changes to the knowledge base while maintaining consistency and correctness. MR imaging is relatively new and hence scanning methods change frequently; also there is a shortage of experts in this area and they often differ in their approaches. Thus an expert system in this area will need extensive maintenance facilities, both to keep the knowledge base up-to-date and to reflect the individual preferences of different experts.

A prototype system has been developed to advise on parameter settings (about fifty) for an MR scan, given the suspected disease(s) and/or patient symptom(s). Maintenance and debugging facilities are provided. The system has been implemented on a Sun 3/60 workstation using Prolog.

The domain knowledge was elicited from the expert and analysed using the KADS methodology. This formed the basis for the knowledge representation scheme and overall design. The resulting system incorporates facilities (a) to make substantial changes to the knowledge base and (b) for exploring the implications of the changes to ensure consistency and correctness of the modified knowledge base.

The system demonstrates the need for careful analysis of the domain knowledge so as to choose a representation scheme which aids maintenance. While the emphasis in this study has been on MR imaging, the analysis and general approach could be applied to other areas.

University of Greenwich
School of Computing and Information Technology
PhD Awarded