The Hybrid Konsiliar is an interdisciplinary project involving the medical profession and computer scientists in which a clinically oriented, hybrid information system for the neurosciences is to be developed. For the computer scientists HYBRIKON presents the possibility of applying and evaluating informatics methods for knowledge modelling and forms of representation using a subset of an extensive medical domain, namely the neurosciences. The project was supported financially from 1992 to 1995 by the project MEDWIS (Knowledge Bases in Medicine) from the German Federal Ministry of Research and Technology. The medical partners were:
The system consists of a symbol-processing and an image-processing component. The symbol-processing component which we developed consists of a knowledge-based system which contains up-to-date knowledge from the neurosciences. Here we are limited, for the time being, to the specializations of our clinical partners, namely information about the higher brain functions: memory, language and motion. Additionally, the knowledge contained in the Hybrid Konsiliar will always have references to the neuroanatomical structure. During the first phase of the project it transpired that all partners had an active interest in modelling a transmitter system. We therefore designated such a module as an extension to the project.
The integration of the picture and symbol-processing components presupposes a common interface. Therefore our system contains a symbolic brain atlas, that can take in knowledge about area hierarchies, neural connections and spatial neighbourhood relations. It is intended that coordinates of the topographical structures of the digital atlas shall be assigned to the topographic units of the symbolic brain atlas, with which the combination is realizable. Functional predicates can be added to the symbolic atlas (a model of the topography of the brain). We differentiate three methodologically separated blocks with the modelling of neuromedical knowledge (see figure modelling in HYBRIKON): axiomatization, model construction and implementation on a computer.
Modelling in HYBRIKON
In the first phase, axiomatization, a theory is established. This starts with the definition of relevant objects - morphological areas, neural connections and functional centers, for example. A composition of interesting relationships between these objects follows on from this. Apart from the neuroanatomical facts, background theory must also be entered. By this we mean all the logical and geometrical circumstances concerning objects and their relationships. Furthermore typical conclusions and thought patterns of neuroscientific experts are part of this background theory. The axiomatization will of course remain incomplete. It concerns the formal description of substantial sections of knowledge about facts and background theory i.e. in first level predicate calculus or in a suitable extension of this. The final product of the axiomatization is therefore a logical theory consisting of facts and background knowledge.
The model construction phase provides the link between axiomatization and implementation on a computer. For the upset theory, a model, or, more precisely, a parameterized model set, must be designed in which the interpretation of the objects and relationships is made in such a way that all formulas contained in the theory are valid according to the semantics of first level predicate calculus. This phase therefore supplies essentially two results:
We have to ensure that not only do the models satsify the axioms of our background theory, but also that the atomic facts in the models are valid. This is the real interface between the fact base, which is represented in a ER model and the background theory. We implement this connection by designing models, which are saturated. Saturated models are those into which one can embed other monomorphic (or structure preserving) models. By coupling the different forms of representation with saturated models we can reuse the 'constant' part of our neuroanatomical knowledge, once we have modelled it. The fact base, which will always be being modified and extended, can be administered more easily using a data base system. It can then be embedded into the saturated models of our background theory; this means that we do not have to redesign the models once they have first been designed.
In the implementation phase the set-theoretical model is transferred to appropriate data structures and procedures using a user friendly programming language. Foe example, an area hierarchy has to be realized by means of a tree-like data structure. The ER model for the description of the fact knowledge is translated into appropriate data base schemata.
In summary, it is to be determined that, by partitioning the modelling into axiomatisation, model construction and implementation, reliable monitoring of the performance of the system is guaranteed. In particular the model construction provides an important connection between the purely logical description and the implementation on a computer. Because it is a consultation system for a complex, extendable and modifiable knowledge area, the development of user-friendly interfaces for the system maintenance, knowldege acquisition and consultation are of crucial importance for the acceptance of the system.
For consultation two alternative access methods are proposed: first, a natural language interface offers users the possibility of formulating their questions using a very limited natural language. These questions are translated with the help of a translator into SQL statements for the data base inquiry or a search program in a user friendly programming language. Secondly, a graphical access component offers users the possibility of using the ER model or a visualized graph model of the symbolic atlas for their inquiries. Since our embedded knowledge always has a topographical reference there is also the possibility of integrating figures from standard atlases.
Following on from the development of this information system and the tools for its maintenance and the acquisition of data, the evaluation and validation of the system pose further problems. For the evaluation and rating of both the modelling and the user interfaces criteria for quality asessment have to be decided. These will require not just the use of general scientific measurements, but also observation of the types of communication between computer scientists and the medical profession in this field of knowledge engineering and the problems that arise in their communications.
Publications: [Brendel, Piron, Widdig 1992], [Brendel, Widdig, Piron, Schinzel 1993], [Brendel, Widdig, Piron, Keyserlingk 1993].
| Project Director: | Prof. Dr. Britta Schinzel |
| Project Participants: | Oliver Brendel, Rolf Widdig, Frank Piron |
| Promotion: | BMFT, 4-92 bis 3-95 |