Skip to main content

A General Architecture for Medical Expert Systems


A suitable general architecture for medical expert systems has been defined in terms of generic tasks (diagnosis, therapy planning and monitoring) to be executed for developing medical reasoning in a patient's management. Their epistemological structure has been investigated. The focus of the project was on the comparison and validation of different approaches in order to implement the generate and test cycle, which represents the basic feature of the epistemological model of medical reasoning developed. The project created a new abstraction paradigm aiming at unifying various ideas for the analysis and design of a knowledge based system (KBS). These ideas all attempt to provide descriptions of KBS reasoning for solving a particular kind of problems at a conceptual level that is above the implementation. The proposal was based on a 2-levels analysis of KBS: an epistemological and a computational level. Ontology and inference model of a KBS have to be defined. Ontology represents the conceptual model of entities and relationships composing the domain knowledge, while the inference model is the conceptual representation of the inference structure to execute a task by managing that ontology. The project identified 3 generic tasks executed by medical KBS: diagnosis, therapy planning and monitoring. They are interrelated but can be submitted to separate epistemological analyses. The main result of these analyses was that these generic tasks manage different ontologies, but can be executed using a unique inference model. Such a model involves 3 different inference types: abduction, deduction, and induction. Thus, medical reasoning may be broken down into 3 different phases: initial information is used to generate (abduce) plausible hypotheses (hypothesis generation phase), then these hypotheses are used as starting conditions to forecast (deduce) expected consequences, then these expected consequences are matched with the state of affairs in the patient in order to confirm or falsify (induce) these hypotheses (hypothesis testing phase).
We propose a coherent project that will address the definition of a General Architecture for Medical Expert Systems, called GAMES. New representations of complex medical knowledge and reasoning processes will be explored. Experimentation at two levels will play a major role : On the one hand, we will develop small expert systems prototypes ; on the other hand, paper and pencil scenarios will serve as a basis for envisaging the main features of new generation medical expert systems.
We believe that medical practice requires the use of categorical and probabilistic reasoning and mathematical modelling, and that many tasks should be addressed by combinations of these approaches. The GAMES project will help the necessary process of cross-fertilizing artificial intelligence and decision theory : It will use categorical reasoning to develop and explain diagnostic conclusions, probabilistic and utility-based reasoning to address the problem of therapy planning, and categorical reasoning and mathematical modelling and case base reasoning to monitor treatment and recommend changes to treatment, if needed.
The clinical value of system modules and methodology tools will be assessed and evaluated with medical experts in the domains of anemia and breast cancer. The objectives of evaluation are twofold :
a) to assess the relevance and reliability of GAMES in the aforementioned clinical domains and
b) to assess the degree to which the structure proposed by GAMES can provide a system platform for developing and implementing KBS's in other clinical domains and which could be integrated with other HIS (Hospital Information Systems) components, too, for improving the process of health-care delivery.
Cross-fertilization opportunities between AIM and other European initiatives will be highlighted as well as requirements for an innovative medical expert systems architecture.
Main Deliverables :
Specification report on system architecture and knowledge representation of the Diagnoser Prototypes of therapy advisor, patient monitor and interfaces in the areas of anemia and breast cancer.


Data not available

Zaproszenie do składania wniosków

Data not available

System finansowania

Data not available


Enidata SpA

Uczestnicy (7)


271 Pavia

Zobacz na mapie

University of Aberdeen
United Kingdom
Regent Walk
AB9 1FX Aberdeen

Zobacz na mapie

Université d'Aix-Marseille II

13288 Marseille

Zobacz na mapie